]> git.proxmox.com Git - mirror_qemu.git/blobdiff - net/vhost-vdpa.c
vdpa: Check device ack in vhost_vdpa_net_load_rx_mode()
[mirror_qemu.git] / net / vhost-vdpa.c
index 0c1c0760a7a164881f45b7583106b215a0557796..36a4e57c0dcd3fdf399e72b87dca982d884f44d5 100644 (file)
@@ -75,11 +75,14 @@ const int vdpa_feature_bits[] = {
     VIRTIO_NET_F_GUEST_TSO4,
     VIRTIO_NET_F_GUEST_TSO6,
     VIRTIO_NET_F_GUEST_UFO,
+    VIRTIO_NET_F_GUEST_USO4,
+    VIRTIO_NET_F_GUEST_USO6,
     VIRTIO_NET_F_HASH_REPORT,
     VIRTIO_NET_F_HOST_ECN,
     VIRTIO_NET_F_HOST_TSO4,
     VIRTIO_NET_F_HOST_TSO6,
     VIRTIO_NET_F_HOST_UFO,
+    VIRTIO_NET_F_HOST_USO,
     VIRTIO_NET_F_MQ,
     VIRTIO_NET_F_MRG_RXBUF,
     VIRTIO_NET_F_MTU,
@@ -111,6 +114,8 @@ static const uint64_t vdpa_svq_device_features =
     BIT_ULL(VIRTIO_NET_F_STATUS) |
     BIT_ULL(VIRTIO_NET_F_CTRL_VQ) |
     BIT_ULL(VIRTIO_NET_F_CTRL_RX) |
+    BIT_ULL(VIRTIO_NET_F_CTRL_VLAN) |
+    BIT_ULL(VIRTIO_NET_F_CTRL_RX_EXTRA) |
     BIT_ULL(VIRTIO_NET_F_MQ) |
     BIT_ULL(VIRTIO_F_ANY_LAYOUT) |
     BIT_ULL(VIRTIO_NET_F_CTRL_MAC_ADDR) |
@@ -370,6 +375,22 @@ static int vhost_vdpa_net_data_start(NetClientState *nc)
     return 0;
 }
 
+static int vhost_vdpa_net_data_load(NetClientState *nc)
+{
+    VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc);
+    struct vhost_vdpa *v = &s->vhost_vdpa;
+    bool has_cvq = v->dev->vq_index_end % 2;
+
+    if (has_cvq) {
+        return 0;
+    }
+
+    for (int i = 0; i < v->dev->nvqs; ++i) {
+        vhost_vdpa_set_vring_ready(v, i + v->dev->vq_index);
+    }
+    return 0;
+}
+
 static void vhost_vdpa_net_client_stop(NetClientState *nc)
 {
     VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc);
@@ -384,6 +405,8 @@ static void vhost_vdpa_net_client_stop(NetClientState *nc)
     dev = s->vhost_vdpa.dev;
     if (dev->vq_index + dev->nvqs == dev->vq_index_end) {
         g_clear_pointer(&s->vhost_vdpa.iova_tree, vhost_iova_tree_delete);
+    } else {
+        s->vhost_vdpa.iova_tree = NULL;
     }
 }
 
@@ -392,6 +415,7 @@ static NetClientInfo net_vhost_vdpa_info = {
         .size = sizeof(VhostVDPAState),
         .receive = vhost_vdpa_receive,
         .start = vhost_vdpa_net_data_start,
+        .load = vhost_vdpa_net_data_load,
         .stop = vhost_vdpa_net_client_stop,
         .cleanup = vhost_vdpa_cleanup,
         .has_vnet_hdr = vhost_vdpa_has_vnet_hdr,
@@ -504,7 +528,7 @@ static int vhost_vdpa_net_cvq_start(NetClientState *nc)
 
     s0 = vhost_vdpa_net_first_nc_vdpa(s);
     v->shadow_data = s0->vhost_vdpa.shadow_vqs_enabled;
-    v->shadow_vqs_enabled = s->always_svq;
+    v->shadow_vqs_enabled = s0->vhost_vdpa.shadow_vqs_enabled;
     s->vhost_vdpa.address_space_id = VHOST_VDPA_GUEST_PA_ASID;
 
     if (s->vhost_vdpa.shadow_data) {
@@ -594,22 +618,14 @@ static void vhost_vdpa_net_cvq_stop(NetClientState *nc)
     vhost_vdpa_net_client_stop(nc);
 }
 
-static ssize_t vhost_vdpa_net_cvq_add(VhostVDPAState *s, size_t out_len,
-                                      size_t in_len)
+static ssize_t vhost_vdpa_net_cvq_add(VhostVDPAState *s,
+                                    const struct iovec *out_sg, size_t out_num,
+                                    const struct iovec *in_sg, size_t in_num)
 {
-    /* Buffers for the device */
-    const struct iovec out = {
-        .iov_base = s->cvq_cmd_out_buffer,
-        .iov_len = out_len,
-    };
-    const struct iovec in = {
-        .iov_base = s->status,
-        .iov_len = sizeof(virtio_net_ctrl_ack),
-    };
     VhostShadowVirtqueue *svq = g_ptr_array_index(s->vhost_vdpa.shadow_vqs, 0);
     int r;
 
-    r = vhost_svq_add(svq, &out, 1, &in, 1, NULL);
+    r = vhost_svq_add(svq, out_sg, out_num, in_sg, in_num, NULL);
     if (unlikely(r != 0)) {
         if (unlikely(r == -ENOSPC)) {
             qemu_log_mask(LOG_GUEST_ERROR, "%s: No space on device queue\n",
@@ -623,7 +639,7 @@ static ssize_t vhost_vdpa_net_cvq_add(VhostVDPAState *s, size_t out_len,
      * descriptor. Also, we need to take the answer before SVQ pulls by itself,
      * when BQL is released
      */
-    return vhost_svq_poll(svq);
+    return vhost_svq_poll(svq, 1);
 }
 
 static ssize_t vhost_vdpa_net_load_cmd(VhostVDPAState *s, uint8_t class,
@@ -635,6 +651,15 @@ static ssize_t vhost_vdpa_net_load_cmd(VhostVDPAState *s, uint8_t class,
         .cmd = cmd,
     };
     size_t data_size = iov_size(data_sg, data_num);
+    /* Buffers for the device */
+    const struct iovec out = {
+        .iov_base = s->cvq_cmd_out_buffer,
+        .iov_len = sizeof(ctrl) + data_size,
+    };
+    const struct iovec in = {
+        .iov_base = s->status,
+        .iov_len = sizeof(*s->status),
+    };
 
     assert(data_size < vhost_vdpa_net_cvq_cmd_page_len() - sizeof(ctrl));
 
@@ -645,8 +670,7 @@ static ssize_t vhost_vdpa_net_load_cmd(VhostVDPAState *s, uint8_t class,
     iov_to_buf(data_sg, data_num, 0,
                s->cvq_cmd_out_buffer + sizeof(ctrl), data_size);
 
-    return vhost_vdpa_net_cvq_add(s, data_size + sizeof(ctrl),
-                                  sizeof(virtio_net_ctrl_ack));
+    return vhost_vdpa_net_cvq_add(s, &out, 1, &in, 1);
 }
 
 static int vhost_vdpa_net_load_mac(VhostVDPAState *s, const VirtIONet *n)
@@ -803,14 +827,24 @@ static int vhost_vdpa_net_load_rx_mode(VhostVDPAState *s,
         .iov_base = &on,
         .iov_len = sizeof(on),
     };
-    return vhost_vdpa_net_load_cmd(s, VIRTIO_NET_CTRL_RX,
-                                   cmd, &data, 1);
+    ssize_t dev_written;
+
+    dev_written = vhost_vdpa_net_load_cmd(s, VIRTIO_NET_CTRL_RX,
+                                          cmd, &data, 1);
+    if (unlikely(dev_written < 0)) {
+        return dev_written;
+    }
+    if (*s->status != VIRTIO_NET_OK) {
+        return -EIO;
+    }
+
+    return 0;
 }
 
 static int vhost_vdpa_net_load_rx(VhostVDPAState *s,
                                   const VirtIONet *n)
 {
-    ssize_t dev_written;
+    ssize_t r;
 
     if (!virtio_vdev_has_feature(&n->parent_obj, VIRTIO_NET_F_CTRL_RX)) {
         return 0;
@@ -820,7 +854,7 @@ static int vhost_vdpa_net_load_rx(VhostVDPAState *s,
      * According to virtio_net_reset(), device turns promiscuous mode
      * on by default.
      *
-     * Addtionally, according to VirtIO standard, "Since there are
+     * Additionally, according to VirtIO standard, "Since there are
      * no guarantees, it can use a hash filter or silently switch to
      * allmulti or promiscuous mode if it is given too many addresses.".
      * QEMU marks `n->mac_table.uni_overflow` if guest sets too many
@@ -835,13 +869,9 @@ static int vhost_vdpa_net_load_rx(VhostVDPAState *s,
      * configuration only at live migration.
      */
     if (!n->mac_table.uni_overflow && !n->promisc) {
-        dev_written = vhost_vdpa_net_load_rx_mode(s,
-                                            VIRTIO_NET_CTRL_RX_PROMISC, 0);
-        if (unlikely(dev_written < 0)) {
-            return dev_written;
-        }
-        if (*s->status != VIRTIO_NET_OK) {
-            return -EIO;
+        r = vhost_vdpa_net_load_rx_mode(s, VIRTIO_NET_CTRL_RX_PROMISC, 0);
+        if (unlikely(r < 0)) {
+            return r;
         }
     }
 
@@ -863,13 +893,9 @@ static int vhost_vdpa_net_load_rx(VhostVDPAState *s,
      * configuration only at live migration.
      */
     if (n->mac_table.multi_overflow || n->allmulti) {
-        dev_written = vhost_vdpa_net_load_rx_mode(s,
-                                            VIRTIO_NET_CTRL_RX_ALLMULTI, 1);
-        if (unlikely(dev_written < 0)) {
-            return dev_written;
-        }
-        if (*s->status != VIRTIO_NET_OK) {
-            return -EIO;
+        r = vhost_vdpa_net_load_rx_mode(s, VIRTIO_NET_CTRL_RX_ALLMULTI, 1);
+        if (unlikely(r < 0)) {
+            return r;
         }
     }
 
@@ -888,13 +914,9 @@ static int vhost_vdpa_net_load_rx(VhostVDPAState *s,
      * configuration only at live migration.
      */
     if (n->alluni) {
-        dev_written = vhost_vdpa_net_load_rx_mode(s,
-                                            VIRTIO_NET_CTRL_RX_ALLUNI, 1);
-        if (dev_written < 0) {
-            return dev_written;
-        }
-        if (*s->status != VIRTIO_NET_OK) {
-            return -EIO;
+        r = vhost_vdpa_net_load_rx_mode(s, VIRTIO_NET_CTRL_RX_ALLUNI, 1);
+        if (r < 0) {
+            return r;
         }
     }
 
@@ -909,13 +931,9 @@ static int vhost_vdpa_net_load_rx(VhostVDPAState *s,
      * configuration only at live migration.
      */
     if (n->nomulti) {
-        dev_written = vhost_vdpa_net_load_rx_mode(s,
-                                            VIRTIO_NET_CTRL_RX_NOMULTI, 1);
-        if (dev_written < 0) {
-            return dev_written;
-        }
-        if (*s->status != VIRTIO_NET_OK) {
-            return -EIO;
+        r = vhost_vdpa_net_load_rx_mode(s, VIRTIO_NET_CTRL_RX_NOMULTI, 1);
+        if (r < 0) {
+            return r;
         }
     }
 
@@ -930,13 +948,9 @@ static int vhost_vdpa_net_load_rx(VhostVDPAState *s,
      * configuration only at live migration.
      */
     if (n->nouni) {
-        dev_written = vhost_vdpa_net_load_rx_mode(s,
-                                            VIRTIO_NET_CTRL_RX_NOUNI, 1);
-        if (dev_written < 0) {
-            return dev_written;
-        }
-        if (*s->status != VIRTIO_NET_OK) {
-            return -EIO;
+        r = vhost_vdpa_net_load_rx_mode(s, VIRTIO_NET_CTRL_RX_NOUNI, 1);
+        if (r < 0) {
+            return r;
         }
     }
 
@@ -951,20 +965,60 @@ static int vhost_vdpa_net_load_rx(VhostVDPAState *s,
      * configuration only at live migration.
      */
     if (n->nobcast) {
-        dev_written = vhost_vdpa_net_load_rx_mode(s,
-                                            VIRTIO_NET_CTRL_RX_NOBCAST, 1);
-        if (dev_written < 0) {
-            return dev_written;
+        r = vhost_vdpa_net_load_rx_mode(s, VIRTIO_NET_CTRL_RX_NOBCAST, 1);
+        if (r < 0) {
+            return r;
         }
-        if (*s->status != VIRTIO_NET_OK) {
-            return -EIO;
+    }
+
+    return 0;
+}
+
+static int vhost_vdpa_net_load_single_vlan(VhostVDPAState *s,
+                                           const VirtIONet *n,
+                                           uint16_t vid)
+{
+    const struct iovec data = {
+        .iov_base = &vid,
+        .iov_len = sizeof(vid),
+    };
+    ssize_t dev_written = vhost_vdpa_net_load_cmd(s, VIRTIO_NET_CTRL_VLAN,
+                                                  VIRTIO_NET_CTRL_VLAN_ADD,
+                                                  &data, 1);
+    if (unlikely(dev_written < 0)) {
+        return dev_written;
+    }
+    if (unlikely(*s->status != VIRTIO_NET_OK)) {
+        return -EIO;
+    }
+
+    return 0;
+}
+
+static int vhost_vdpa_net_load_vlan(VhostVDPAState *s,
+                                    const VirtIONet *n)
+{
+    int r;
+
+    if (!virtio_vdev_has_feature(&n->parent_obj, VIRTIO_NET_F_CTRL_VLAN)) {
+        return 0;
+    }
+
+    for (int i = 0; i < MAX_VLAN >> 5; i++) {
+        for (int j = 0; n->vlans[i] && j <= 0x1f; j++) {
+            if (n->vlans[i] & (1U << j)) {
+                r = vhost_vdpa_net_load_single_vlan(s, n, (i << 5) + j);
+                if (unlikely(r != 0)) {
+                    return r;
+                }
+            }
         }
     }
 
     return 0;
 }
 
-static int vhost_vdpa_net_load(NetClientState *nc)
+static int vhost_vdpa_net_cvq_load(NetClientState *nc)
 {
     VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc);
     struct vhost_vdpa *v = &s->vhost_vdpa;
@@ -973,26 +1027,34 @@ static int vhost_vdpa_net_load(NetClientState *nc)
 
     assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA);
 
-    if (!v->shadow_vqs_enabled) {
-        return 0;
-    }
+    vhost_vdpa_set_vring_ready(v, v->dev->vq_index);
 
-    n = VIRTIO_NET(v->dev->vdev);
-    r = vhost_vdpa_net_load_mac(s, n);
-    if (unlikely(r < 0)) {
-        return r;
-    }
-    r = vhost_vdpa_net_load_mq(s, n);
-    if (unlikely(r)) {
-        return r;
-    }
-    r = vhost_vdpa_net_load_offloads(s, n);
-    if (unlikely(r)) {
-        return r;
+    if (v->shadow_vqs_enabled) {
+        n = VIRTIO_NET(v->dev->vdev);
+        r = vhost_vdpa_net_load_mac(s, n);
+        if (unlikely(r < 0)) {
+            return r;
+        }
+        r = vhost_vdpa_net_load_mq(s, n);
+        if (unlikely(r)) {
+            return r;
+        }
+        r = vhost_vdpa_net_load_offloads(s, n);
+        if (unlikely(r)) {
+            return r;
+        }
+        r = vhost_vdpa_net_load_rx(s, n);
+        if (unlikely(r)) {
+            return r;
+        }
+        r = vhost_vdpa_net_load_vlan(s, n);
+        if (unlikely(r)) {
+            return r;
+        }
     }
-    r = vhost_vdpa_net_load_rx(s, n);
-    if (unlikely(r)) {
-        return r;
+
+    for (int i = 0; i < v->dev->vq_index; ++i) {
+        vhost_vdpa_set_vring_ready(v, i);
     }
 
     return 0;
@@ -1003,7 +1065,7 @@ static NetClientInfo net_vhost_vdpa_cvq_info = {
     .size = sizeof(VhostVDPAState),
     .receive = vhost_vdpa_receive,
     .start = vhost_vdpa_net_cvq_start,
-    .load = vhost_vdpa_net_load,
+    .load = vhost_vdpa_net_cvq_load,
     .stop = vhost_vdpa_net_cvq_stop,
     .cleanup = vhost_vdpa_cleanup,
     .has_vnet_hdr = vhost_vdpa_has_vnet_hdr,
@@ -1038,12 +1100,14 @@ static NetClientInfo net_vhost_vdpa_cvq_info = {
  */
 static int vhost_vdpa_net_excessive_mac_filter_cvq_add(VhostVDPAState *s,
                                                        VirtQueueElement *elem,
-                                                       struct iovec *out)
+                                                       struct iovec *out,
+                                                       const struct iovec *in)
 {
     struct virtio_net_ctrl_mac mac_data, *mac_ptr;
     struct virtio_net_ctrl_hdr *hdr_ptr;
     uint32_t cursor;
     ssize_t r;
+    uint8_t on = 1;
 
     /* parse the non-multicast MAC address entries from CVQ command */
     cursor = sizeof(*hdr_ptr);
@@ -1091,7 +1155,13 @@ static int vhost_vdpa_net_excessive_mac_filter_cvq_add(VhostVDPAState *s,
      * filter table to the vdpa device, it should send the
      * VIRTIO_NET_CTRL_RX_PROMISC CVQ command to enable promiscuous mode
      */
-    r = vhost_vdpa_net_load_rx_mode(s, VIRTIO_NET_CTRL_RX_PROMISC, 1);
+    hdr_ptr = out->iov_base;
+    out->iov_len = sizeof(*hdr_ptr) + sizeof(on);
+
+    hdr_ptr->class = VIRTIO_NET_CTRL_RX;
+    hdr_ptr->cmd = VIRTIO_NET_CTRL_RX_PROMISC;
+    iov_from_buf(out, 1, sizeof(*hdr_ptr), &on, sizeof(on));
+    r = vhost_vdpa_net_cvq_add(s, out, 1, in, 1);
     if (unlikely(r < 0)) {
         return r;
     }
@@ -1129,7 +1199,7 @@ static int vhost_vdpa_net_excessive_mac_filter_cvq_add(VhostVDPAState *s,
      * Pack the non-multicast MAC addresses part for fake CVQ command.
      *
      * According to virtio_net_handle_mac(), QEMU doesn't verify the MAC
-     * addresses provieded in CVQ command. Therefore, only the entries
+     * addresses provided in CVQ command. Therefore, only the entries
      * field need to be prepared in the CVQ command.
      */
     mac_ptr = out->iov_base + cursor;
@@ -1140,7 +1210,7 @@ static int vhost_vdpa_net_excessive_mac_filter_cvq_add(VhostVDPAState *s,
      * Pack the multicast MAC addresses part for fake CVQ command.
      *
      * According to virtio_net_handle_mac(), QEMU doesn't verify the MAC
-     * addresses provieded in CVQ command. Therefore, only the entries
+     * addresses provided in CVQ command. Therefore, only the entries
      * field need to be prepared in the CVQ command.
      */
     mac_ptr = out->iov_base + cursor;
@@ -1172,10 +1242,15 @@ static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq,
         .iov_base = s->cvq_cmd_out_buffer,
     };
     /* in buffer used for device model */
-    const struct iovec in = {
+    const struct iovec model_in = {
         .iov_base = &status,
         .iov_len = sizeof(status),
     };
+    /* in buffer used for vdpa device */
+    const struct iovec vdpa_in = {
+        .iov_base = s->status,
+        .iov_len = sizeof(*s->status),
+    };
     ssize_t dev_written = -EINVAL;
 
     out.iov_len = iov_to_buf(elem->out_sg, elem->out_num, 0,
@@ -1201,15 +1276,15 @@ static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq,
          * rejects the flawed CVQ command.
          *
          * Therefore, QEMU must handle this situation instead of sending
-         * the CVQ command direclty.
+         * the CVQ command directly.
          */
         dev_written = vhost_vdpa_net_excessive_mac_filter_cvq_add(s, elem,
-                                                                  &out);
+                                                            &out, &vdpa_in);
         if (unlikely(dev_written < 0)) {
             goto out;
         }
     } else {
-        dev_written = vhost_vdpa_net_cvq_add(s, out.iov_len, sizeof(status));
+        dev_written = vhost_vdpa_net_cvq_add(s, &out, 1, &vdpa_in, 1);
         if (unlikely(dev_written < 0)) {
             goto out;
         }
@@ -1225,7 +1300,7 @@ static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq,
     }
 
     status = VIRTIO_NET_ERR;
-    virtio_net_handle_ctrl_iov(svq->vdev, &in, 1, &out, 1);
+    virtio_net_handle_ctrl_iov(svq->vdev, &model_in, 1, &out, 1);
     if (status != VIRTIO_NET_OK) {
         error_report("Bad CVQ processing in model");
     }
@@ -1269,8 +1344,7 @@ static int vhost_vdpa_probe_cvq_isolation(int device_fd, uint64_t features,
     uint64_t backend_features;
     int64_t cvq_group;
     uint8_t status = VIRTIO_CONFIG_S_ACKNOWLEDGE |
-                     VIRTIO_CONFIG_S_DRIVER |
-                     VIRTIO_CONFIG_S_FEATURES_OK;
+                     VIRTIO_CONFIG_S_DRIVER;
     int r;
 
     ERRP_GUARD();
@@ -1285,14 +1359,22 @@ static int vhost_vdpa_probe_cvq_isolation(int device_fd, uint64_t features,
         return 0;
     }
 
+    r = ioctl(device_fd, VHOST_VDPA_SET_STATUS, &status);
+    if (unlikely(r)) {
+        error_setg_errno(errp, -r, "Cannot set device status");
+        goto out;
+    }
+
     r = ioctl(device_fd, VHOST_SET_FEATURES, &features);
     if (unlikely(r)) {
-        error_setg_errno(errp, errno, "Cannot set features");
+        error_setg_errno(errp, -r, "Cannot set features");
+        goto out;
     }
 
+    status |= VIRTIO_CONFIG_S_FEATURES_OK;
     r = ioctl(device_fd, VHOST_VDPA_SET_STATUS, &status);
     if (unlikely(r)) {
-        error_setg_errno(errp, -r, "Cannot set device features");
+        error_setg_errno(errp, -r, "Cannot set device status");
         goto out;
     }
 
@@ -1351,7 +1433,7 @@ static NetClientState *net_vhost_vdpa_init(NetClientState *peer,
     VhostVDPAState *s;
     int ret = 0;
     assert(name);
-    int cvq_isolated;
+    int cvq_isolated = 0;
 
     if (is_datapath) {
         nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device,
@@ -1391,18 +1473,6 @@ static NetClientState *net_vhost_vdpa_init(NetClientState *peer,
         s->vhost_vdpa.shadow_vq_ops = &vhost_vdpa_net_svq_ops;
         s->vhost_vdpa.shadow_vq_ops_opaque = s;
         s->cvq_isolated = cvq_isolated;
-
-        /*
-         * TODO: We cannot migrate devices with CVQ and no x-svq enabled as
-         * there is no way to set the device state (MAC, MQ, etc) before
-         * starting the datapath.
-         *
-         * Migration blocker ownership now belongs to s->vhost_vdpa.
-         */
-        if (!svq) {
-            error_setg(&s->vhost_vdpa.migration_blocker,
-                       "net vdpa cannot migrate with CVQ feature");
-        }
     }
     ret = vhost_vdpa_add(nc, (void *)&s->vhost_vdpa, queue_pair_index, nvqs);
     if (ret) {