examples/vhost: demonstrate the new generic APIs
authorYuanhan Liu <yuanhan.liu@linux.intel.com>
Sat, 1 Apr 2017 07:23:00 +0000 (15:23 +0800)
committerYuanhan Liu <yuanhan.liu@linux.intel.com>
Sat, 1 Apr 2017 08:49:47 +0000 (10:49 +0200)
Now DPDK vhost lib has been generic enough, that it can be used to
implement any vhost-user drivers.

For example, this patch implements a very simple vhost-user net driver,
mainly for demonstrating how to use those generic vhost APIs.

And when the --builtin-net-driver option is used, the example virtio-net
driver code will be invoked, instead of the one provided from the vhost
library.

Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
examples/vhost/Makefile
examples/vhost/main.c
examples/vhost/main.h
examples/vhost/virtio_net.c [new file with mode: 0644]

index e95c68a..af7be99 100644 (file)
@@ -48,7 +48,7 @@ else
 APP = vhost-switch
 
 # all source are stored in SRCS-y
-SRCS-y := main.c
+SRCS-y := main.c virtio_net.c
 
 CFLAGS += -O2 -D_FILE_OFFSET_BITS=64
 CFLAGS += $(WERROR_FLAGS)
index 08b82f6..022438e 100644 (file)
@@ -1,7 +1,7 @@
 /*-
  *   BSD LICENSE
  *
- *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
+ *   Copyright(c) 2010-2017 Intel Corporation. All rights reserved.
  *   All rights reserved.
  *
  *   Redistribution and use in source and binary forms, with or without
@@ -65,7 +65,6 @@
 #define MBUF_CACHE_SIZE        128
 #define MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE
 
-#define MAX_PKT_BURST 32               /* Max burst size for RX/TX */
 #define BURST_TX_DRAIN_US 100  /* TX drain every ~100us */
 
 #define BURST_RX_WAIT_US 15    /* Defines how long we wait between retries on RX */
@@ -129,6 +128,8 @@ static uint32_t enable_tso;
 static int client_mode;
 static int dequeue_zero_copy;
 
+static int builtin_net_driver;
+
 /* Specify timeout (in useconds) between retries on RX. */
 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
 /* Specify the number of retries on RX. */
@@ -499,6 +500,7 @@ us_vhost_parse_args(int argc, char **argv)
                {"tso", required_argument, NULL, 0},
                {"client", no_argument, &client_mode, 1},
                {"dequeue-zero-copy", no_argument, &dequeue_zero_copy, 1},
+               {"builtin-net-driver", no_argument, &builtin_net_driver, 1},
                {NULL, 0, 0, 0},
        };
 
@@ -795,7 +797,12 @@ virtio_xmit(struct vhost_dev *dst_vdev, struct vhost_dev *src_vdev,
 {
        uint16_t ret;
 
-       ret = rte_vhost_enqueue_burst(dst_vdev->vid, VIRTIO_RXQ, &m, 1);
+       if (builtin_net_driver) {
+               ret = vs_enqueue_pkts(dst_vdev, VIRTIO_RXQ, &m, 1);
+       } else {
+               ret = rte_vhost_enqueue_burst(dst_vdev->vid, VIRTIO_RXQ, &m, 1);
+       }
+
        if (enable_stats) {
                rte_atomic64_inc(&dst_vdev->stats.rx_total_atomic);
                rte_atomic64_add(&dst_vdev->stats.rx_atomic, ret);
@@ -1066,8 +1073,13 @@ drain_eth_rx(struct vhost_dev *vdev)
                }
        }
 
-       enqueue_count = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ,
+       if (builtin_net_driver) {
+               enqueue_count = vs_enqueue_pkts(vdev, VIRTIO_RXQ,
                                                pkts, rx_count);
+       } else {
+               enqueue_count = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ,
+                                               pkts, rx_count);
+       }
        if (enable_stats) {
                rte_atomic64_add(&vdev->stats.rx_total_atomic, rx_count);
                rte_atomic64_add(&vdev->stats.rx_atomic, enqueue_count);
@@ -1083,8 +1095,13 @@ drain_virtio_tx(struct vhost_dev *vdev)
        uint16_t count;
        uint16_t i;
 
-       count = rte_vhost_dequeue_burst(vdev->vid, VIRTIO_TXQ, mbuf_pool,
+       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);
+       }
 
        /* setup VMDq for the first packet */
        if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && count) {
@@ -1187,6 +1204,9 @@ destroy_device(int vid)
                rte_pause();
        }
 
+       if (builtin_net_driver)
+               vs_vhost_net_remove(vdev);
+
        TAILQ_REMOVE(&lcore_info[vdev->coreid].vdev_list, vdev,
                     lcore_vdev_entry);
        TAILQ_REMOVE(&vhost_dev_list, vdev, global_vdev_entry);
@@ -1235,6 +1255,9 @@ new_device(int vid)
        }
        vdev->vid = vid;
 
+       if (builtin_net_driver)
+               vs_vhost_net_setup(vdev);
+
        TAILQ_INSERT_TAIL(&vhost_dev_list, vdev, global_vdev_entry);
        vdev->vmdq_rx_q = vid * queues_per_pool + vmdq_queue_base;
 
@@ -1513,6 +1536,10 @@ main(int argc, char *argv[])
                        rte_exit(EXIT_FAILURE,
                                "vhost driver register failure.\n");
                }
+
+               if (builtin_net_driver)
+                       rte_vhost_driver_set_features(file, VIRTIO_NET_FEATURES);
+
                if (mergeable == 0) {
                        rte_vhost_driver_disable_features(file,
                                1ULL << VIRTIO_NET_F_MRG_RXBUF);
index ddcd858..9a2aca3 100644 (file)
@@ -1,7 +1,7 @@
 /*-
  *   BSD LICENSE
  *
- *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
+ *   Copyright(c) 2010-2017 Intel Corporation. All rights reserved.
  *   All rights reserved.
  *
  *   Redistribution and use in source and binary forms, with or without
@@ -45,6 +45,8 @@
 
 enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM};
 
+#define MAX_PKT_BURST 32               /* Max burst size for RX/TX */
+
 struct device_statistics {
        uint64_t        tx;
        uint64_t        tx_total;
@@ -52,6 +54,12 @@ struct device_statistics {
        rte_atomic64_t  rx_total_atomic;
 };
 
+struct vhost_queue {
+       struct rte_vhost_vring  vr;
+       uint16_t                last_avail_idx;
+       uint16_t                last_used_idx;
+};
+
 struct vhost_dev {
        /**< Number of memory regions for gpa to hpa translation. */
        uint32_t nregions_hpa;
@@ -69,9 +77,16 @@ struct vhost_dev {
        volatile uint8_t remove;
 
        int vid;
+       uint64_t features;
+       size_t hdr_len;
+       uint16_t nr_vrings;
+       struct rte_vhost_memory *mem;
        struct device_statistics stats;
        TAILQ_ENTRY(vhost_dev) global_vdev_entry;
        TAILQ_ENTRY(vhost_dev) lcore_vdev_entry;
+
+#define MAX_QUEUE_PAIRS        4
+       struct vhost_queue queues[MAX_QUEUE_PAIRS * 2];
 } __rte_cache_aligned;
 
 TAILQ_HEAD(vhost_dev_tailq_list, vhost_dev);
@@ -92,4 +107,15 @@ struct lcore_info {
        struct vhost_dev_tailq_list vdev_list;
 };
 
+/* we implement non-extra virtio net features */
+#define VIRTIO_NET_FEATURES    0
+
+void vs_vhost_net_setup(struct vhost_dev *dev);
+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);
 #endif /* _MAIN_H_ */
diff --git a/examples/vhost/virtio_net.c b/examples/vhost/virtio_net.c
new file mode 100644 (file)
index 0000000..cc2c3d8
--- /dev/null
@@ -0,0 +1,403 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2010-2017 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <stdint.h>
+#include <stdbool.h>
+#include <linux/virtio_net.h>
+
+#include <rte_mbuf.h>
+#include <rte_memcpy.h>
+#include <rte_vhost.h>
+
+#include "main.h"
+
+/*
+ * A very simple vhost-user net driver implementation, without
+ * any extra features being enabled, such as TSO and mrg-Rx.
+ */
+
+void
+vs_vhost_net_setup(struct vhost_dev *dev)
+{
+       uint16_t i;
+       int vid = dev->vid;
+       struct vhost_queue *queue;
+
+       RTE_LOG(INFO, VHOST_CONFIG,
+               "setting builtin vhost-user net driver\n");
+
+       rte_vhost_get_negotiated_features(vid, &dev->features);
+       if (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF))
+               dev->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
+       else
+               dev->hdr_len = sizeof(struct virtio_net_hdr);
+
+       rte_vhost_get_mem_table(vid, &dev->mem);
+
+       dev->nr_vrings = rte_vhost_get_vring_num(vid);
+       for (i = 0; i < dev->nr_vrings; i++) {
+               queue = &dev->queues[i];
+
+               queue->last_used_idx  = 0;
+               queue->last_avail_idx = 0;
+               rte_vhost_get_vhost_vring(vid, i, &queue->vr);
+       }
+}
+
+void
+vs_vhost_net_remove(struct vhost_dev *dev)
+{
+       free(dev->mem);
+}
+
+static inline int __attribute__((always_inline))
+enqueue_pkt(struct vhost_dev *dev, struct rte_vhost_vring *vr,
+           struct rte_mbuf *m, uint16_t desc_idx)
+{
+       uint32_t desc_avail, desc_offset;
+       uint32_t mbuf_avail, mbuf_offset;
+       uint32_t cpy_len;
+       struct vring_desc *desc;
+       uint64_t desc_addr;
+       struct virtio_net_hdr virtio_hdr = {0, 0, 0, 0, 0, 0};
+       /* A counter to avoid desc dead loop chain */
+       uint16_t nr_desc = 1;
+
+       desc = &vr->desc[desc_idx];
+       desc_addr = rte_vhost_gpa_to_vva(dev->mem, desc->addr);
+       /*
+        * Checking of 'desc_addr' placed outside of 'unlikely' macro to avoid
+        * performance issue with some versions of gcc (4.8.4 and 5.3.0) which
+        * otherwise stores offset on the stack instead of in a register.
+        */
+       if (unlikely(desc->len < dev->hdr_len) || !desc_addr)
+               return -1;
+
+       rte_prefetch0((void *)(uintptr_t)desc_addr);
+
+       /* write virtio-net header */
+       *(struct virtio_net_hdr *)(uintptr_t)desc_addr = virtio_hdr;
+
+       desc_offset = dev->hdr_len;
+       desc_avail  = desc->len - dev->hdr_len;
+
+       mbuf_avail  = rte_pktmbuf_data_len(m);
+       mbuf_offset = 0;
+       while (mbuf_avail != 0 || m->next != NULL) {
+               /* done with current mbuf, fetch next */
+               if (mbuf_avail == 0) {
+                       m = m->next;
+
+                       mbuf_offset = 0;
+                       mbuf_avail  = rte_pktmbuf_data_len(m);
+               }
+
+               /* done with current desc buf, fetch next */
+               if (desc_avail == 0) {
+                       if ((desc->flags & VRING_DESC_F_NEXT) == 0) {
+                               /* Room in vring buffer is not enough */
+                               return -1;
+                       }
+                       if (unlikely(desc->next >= vr->size ||
+                                    ++nr_desc > vr->size))
+                               return -1;
+
+                       desc = &vr->desc[desc->next];
+                       desc_addr = rte_vhost_gpa_to_vva(dev->mem, desc->addr);
+                       if (unlikely(!desc_addr))
+                               return -1;
+
+                       desc_offset = 0;
+                       desc_avail  = desc->len;
+               }
+
+               cpy_len = RTE_MIN(desc_avail, mbuf_avail);
+               rte_memcpy((void *)((uintptr_t)(desc_addr + desc_offset)),
+                       rte_pktmbuf_mtod_offset(m, void *, mbuf_offset),
+                       cpy_len);
+
+               mbuf_avail  -= cpy_len;
+               mbuf_offset += cpy_len;
+               desc_avail  -= cpy_len;
+               desc_offset += cpy_len;
+       }
+
+       return 0;
+}
+
+uint16_t
+vs_enqueue_pkts(struct vhost_dev *dev, uint16_t queue_id,
+               struct rte_mbuf **pkts, uint32_t count)
+{
+       struct vhost_queue *queue;
+       struct rte_vhost_vring *vr;
+       uint16_t avail_idx, free_entries, start_idx;
+       uint16_t desc_indexes[MAX_PKT_BURST];
+       uint16_t used_idx;
+       uint32_t i;
+
+       queue = &dev->queues[queue_id];
+       vr    = &queue->vr;
+
+       avail_idx = *((volatile uint16_t *)&vr->avail->idx);
+       start_idx = queue->last_used_idx;
+       free_entries = avail_idx - start_idx;
+       count = RTE_MIN(count, free_entries);
+       count = RTE_MIN(count, (uint32_t)MAX_PKT_BURST);
+       if (count == 0)
+               return 0;
+
+       /* Retrieve all of the desc indexes first to avoid caching issues. */
+       rte_prefetch0(&vr->avail->ring[start_idx & (vr->size - 1)]);
+       for (i = 0; i < count; i++) {
+               used_idx = (start_idx + i) & (vr->size - 1);
+               desc_indexes[i] = vr->avail->ring[used_idx];
+               vr->used->ring[used_idx].id = desc_indexes[i];
+               vr->used->ring[used_idx].len = pkts[i]->pkt_len +
+                                              dev->hdr_len;
+       }
+
+       rte_prefetch0(&vr->desc[desc_indexes[0]]);
+       for (i = 0; i < count; i++) {
+               uint16_t desc_idx = desc_indexes[i];
+               int err;
+
+               err = enqueue_pkt(dev, vr, pkts[i], desc_idx);
+               if (unlikely(err)) {
+                       used_idx = (start_idx + i) & (vr->size - 1);
+                       vr->used->ring[used_idx].len = dev->hdr_len;
+               }
+
+               if (i + 1 < count)
+                       rte_prefetch0(&vr->desc[desc_indexes[i+1]]);
+       }
+
+       rte_smp_wmb();
+
+       *(volatile uint16_t *)&vr->used->idx += count;
+       queue->last_used_idx += count;
+
+       /* flush used->idx update before we read avail->flags. */
+       rte_mb();
+
+       /* Kick the guest if necessary. */
+       if (!(vr->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
+                       && (vr->callfd >= 0))
+               eventfd_write(vr->callfd, (eventfd_t)1);
+       return count;
+}
+
+static inline int __attribute__((always_inline))
+dequeue_pkt(struct vhost_dev *dev, struct rte_vhost_vring *vr,
+           struct rte_mbuf *m, uint16_t desc_idx,
+           struct rte_mempool *mbuf_pool)
+{
+       struct vring_desc *desc;
+       uint64_t desc_addr;
+       uint32_t desc_avail, desc_offset;
+       uint32_t mbuf_avail, mbuf_offset;
+       uint32_t cpy_len;
+       struct rte_mbuf *cur = m, *prev = m;
+       /* A counter to avoid desc dead loop chain */
+       uint32_t nr_desc = 1;
+
+       desc = &vr->desc[desc_idx];
+       if (unlikely((desc->len < dev->hdr_len)) ||
+                       (desc->flags & VRING_DESC_F_INDIRECT))
+               return -1;
+
+       desc_addr = rte_vhost_gpa_to_vva(dev->mem, desc->addr);
+       if (unlikely(!desc_addr))
+               return -1;
+
+       /*
+        * We don't support ANY_LAYOUT, neither VERSION_1, meaning
+        * a Tx packet from guest must have 2 desc buffers at least:
+        * the first for storing the header and the others for
+        * storing the data.
+        *
+        * And since we don't support TSO, we could simply skip the
+        * header.
+        */
+       desc = &vr->desc[desc->next];
+       desc_addr = rte_vhost_gpa_to_vva(dev->mem, desc->addr);
+       if (unlikely(!desc_addr))
+               return -1;
+       rte_prefetch0((void *)(uintptr_t)desc_addr);
+
+       desc_offset = 0;
+       desc_avail  = desc->len;
+       nr_desc    += 1;
+
+       mbuf_offset = 0;
+       mbuf_avail  = m->buf_len - RTE_PKTMBUF_HEADROOM;
+       while (1) {
+               cpy_len = RTE_MIN(desc_avail, mbuf_avail);
+               rte_memcpy(rte_pktmbuf_mtod_offset(cur, void *,
+                                                  mbuf_offset),
+                       (void *)((uintptr_t)(desc_addr + desc_offset)),
+                       cpy_len);
+
+               mbuf_avail  -= cpy_len;
+               mbuf_offset += cpy_len;
+               desc_avail  -= cpy_len;
+               desc_offset += cpy_len;
+
+               /* This desc reaches to its end, get the next one */
+               if (desc_avail == 0) {
+                       if ((desc->flags & VRING_DESC_F_NEXT) == 0)
+                               break;
+
+                       if (unlikely(desc->next >= vr->size ||
+                                    ++nr_desc > vr->size))
+                               return -1;
+                       desc = &vr->desc[desc->next];
+
+                       desc_addr = rte_vhost_gpa_to_vva(dev->mem, desc->addr);
+                       if (unlikely(!desc_addr))
+                               return -1;
+                       rte_prefetch0((void *)(uintptr_t)desc_addr);
+
+                       desc_offset = 0;
+                       desc_avail  = desc->len;
+               }
+
+               /*
+                * This mbuf reaches to its end, get a new one
+                * to hold more data.
+                */
+               if (mbuf_avail == 0) {
+                       cur = rte_pktmbuf_alloc(mbuf_pool);
+                       if (unlikely(cur == NULL)) {
+                               RTE_LOG(ERR, VHOST_DATA, "Failed to "
+                                       "allocate memory for mbuf.\n");
+                               return -1;
+                       }
+
+                       prev->next = cur;
+                       prev->data_len = mbuf_offset;
+                       m->nb_segs += 1;
+                       m->pkt_len += mbuf_offset;
+                       prev = cur;
+
+                       mbuf_offset = 0;
+                       mbuf_avail  = cur->buf_len - RTE_PKTMBUF_HEADROOM;
+               }
+       }
+
+       prev->data_len = mbuf_offset;
+       m->pkt_len    += mbuf_offset;
+
+       return 0;
+}
+
+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)
+{
+       struct vhost_queue *queue;
+       struct rte_vhost_vring *vr;
+       uint32_t desc_indexes[MAX_PKT_BURST];
+       uint32_t used_idx;
+       uint32_t i = 0;
+       uint16_t free_entries;
+       uint16_t avail_idx;
+
+       queue = &dev->queues[queue_id];
+       vr    = &queue->vr;
+
+       free_entries = *((volatile uint16_t *)&vr->avail->idx) -
+                       queue->last_avail_idx;
+       if (free_entries == 0)
+               return 0;
+
+       /* Prefetch available and used ring */
+       avail_idx = queue->last_avail_idx & (vr->size - 1);
+       used_idx  = queue->last_used_idx  & (vr->size - 1);
+       rte_prefetch0(&vr->avail->ring[avail_idx]);
+       rte_prefetch0(&vr->used->ring[used_idx]);
+
+       count = RTE_MIN(count, MAX_PKT_BURST);
+       count = RTE_MIN(count, free_entries);
+
+       /*
+        * Retrieve all of the head indexes first and pre-update used entries
+        * to avoid caching issues.
+        */
+       for (i = 0; i < count; i++) {
+               avail_idx = (queue->last_avail_idx + i) & (vr->size - 1);
+               used_idx  = (queue->last_used_idx  + i) & (vr->size - 1);
+               desc_indexes[i] = vr->avail->ring[avail_idx];
+
+               vr->used->ring[used_idx].id  = desc_indexes[i];
+               vr->used->ring[used_idx].len = 0;
+       }
+
+       /* Prefetch descriptor index. */
+       rte_prefetch0(&vr->desc[desc_indexes[0]]);
+       for (i = 0; i < count; i++) {
+               int err;
+
+               if (likely(i + 1 < count))
+                       rte_prefetch0(&vr->desc[desc_indexes[i + 1]]);
+
+               pkts[i] = rte_pktmbuf_alloc(mbuf_pool);
+               if (unlikely(pkts[i] == NULL)) {
+                       RTE_LOG(ERR, VHOST_DATA,
+                               "Failed to allocate memory for mbuf.\n");
+                       break;
+               }
+
+               err = dequeue_pkt(dev, vr, pkts[i], desc_indexes[i], mbuf_pool);
+               if (unlikely(err)) {
+                       rte_pktmbuf_free(pkts[i]);
+                       break;
+               }
+
+       }
+       if (!i)
+               return 0;
+
+       queue->last_avail_idx += i;
+       queue->last_used_idx += i;
+       rte_smp_wmb();
+       rte_smp_rmb();
+
+       vr->used->idx += i;
+
+       if (!(vr->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
+                       && (vr->callfd >= 0))
+               eventfd_write(vr->callfd, (eventfd_t)1);
+
+       return i;
+}