vhost: move vhost-net zerocopy fields to net.c
authorAsias He <asias@redhat.com>
Sat, 27 Apr 2013 07:07:46 +0000 (15:07 +0800)
committerMichael S. Tsirkin <mst@redhat.com>
Wed, 1 May 2013 07:02:52 +0000 (10:02 +0300)
On top of 'vhost: Allow device specific fields per vq', we can move device
specific fields to device virt queue from vhost virt queue.

Signed-off-by: Asias He <asias@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
drivers/vhost/net.c
drivers/vhost/vhost.c
drivers/vhost/vhost.h

index 176aa030dc5fad3796309e6a722108c8c97cc470..8672e0538d59f5af76ad6ab621002962379a76d2 100644 (file)
@@ -64,8 +64,24 @@ enum {
        VHOST_NET_VQ_MAX = 2,
 };
 
+struct vhost_ubuf_ref {
+       struct kref kref;
+       wait_queue_head_t wait;
+       struct vhost_virtqueue *vq;
+};
+
 struct vhost_net_virtqueue {
        struct vhost_virtqueue vq;
+       /* vhost zerocopy support fields below: */
+       /* last used idx for outstanding DMA zerocopy buffers */
+       int upend_idx;
+       /* first used idx for DMA done zerocopy buffers */
+       int done_idx;
+       /* an array of userspace buffers info */
+       struct ubuf_info *ubuf_info;
+       /* Reference counting for outstanding ubufs.
+        * Protected by vq mutex. Writers must also take device mutex. */
+       struct vhost_ubuf_ref *ubufs;
 };
 
 struct vhost_net {
@@ -82,6 +98,88 @@ struct vhost_net {
        bool tx_flush;
 };
 
+static unsigned vhost_zcopy_mask __read_mostly;
+
+void vhost_enable_zcopy(int vq)
+{
+       vhost_zcopy_mask |= 0x1 << vq;
+}
+
+static void vhost_zerocopy_done_signal(struct kref *kref)
+{
+       struct vhost_ubuf_ref *ubufs = container_of(kref, struct vhost_ubuf_ref,
+                                                   kref);
+       wake_up(&ubufs->wait);
+}
+
+struct vhost_ubuf_ref *vhost_ubuf_alloc(struct vhost_virtqueue *vq,
+                                       bool zcopy)
+{
+       struct vhost_ubuf_ref *ubufs;
+       /* No zero copy backend? Nothing to count. */
+       if (!zcopy)
+               return NULL;
+       ubufs = kmalloc(sizeof(*ubufs), GFP_KERNEL);
+       if (!ubufs)
+               return ERR_PTR(-ENOMEM);
+       kref_init(&ubufs->kref);
+       init_waitqueue_head(&ubufs->wait);
+       ubufs->vq = vq;
+       return ubufs;
+}
+
+void vhost_ubuf_put(struct vhost_ubuf_ref *ubufs)
+{
+       kref_put(&ubufs->kref, vhost_zerocopy_done_signal);
+}
+
+void vhost_ubuf_put_and_wait(struct vhost_ubuf_ref *ubufs)
+{
+       kref_put(&ubufs->kref, vhost_zerocopy_done_signal);
+       wait_event(ubufs->wait, !atomic_read(&ubufs->kref.refcount));
+       kfree(ubufs);
+}
+
+int vhost_net_set_ubuf_info(struct vhost_net *n)
+{
+       bool zcopy;
+       int i;
+
+       for (i = 0; i < n->dev.nvqs; ++i) {
+               zcopy = vhost_zcopy_mask & (0x1 << i);
+               if (!zcopy)
+                       continue;
+               n->vqs[i].ubuf_info = kmalloc(sizeof(*n->vqs[i].ubuf_info) *
+                                             UIO_MAXIOV, GFP_KERNEL);
+               if  (!n->vqs[i].ubuf_info)
+                       goto err;
+       }
+       return 0;
+
+err:
+       while (i--) {
+               zcopy = vhost_zcopy_mask & (0x1 << i);
+               if (!zcopy)
+                       continue;
+               kfree(n->vqs[i].ubuf_info);
+       }
+       return -ENOMEM;
+}
+
+void vhost_net_reset_ubuf_info(struct vhost_net *n)
+{
+       int i;
+
+       for (i = 0; i < VHOST_NET_VQ_MAX; i++) {
+               n->vqs[i].done_idx = 0;
+               n->vqs[i].upend_idx = 0;
+               n->vqs[i].ubufs = NULL;
+               kfree(n->vqs[i].ubuf_info);
+               n->vqs[i].ubuf_info = NULL;
+       }
+
+}
+
 static void vhost_net_tx_packet(struct vhost_net *net)
 {
        ++net->tx_packets;
@@ -157,10 +255,12 @@ static void copy_iovec_hdr(const struct iovec *from, struct iovec *to,
 static int vhost_zerocopy_signal_used(struct vhost_net *net,
                                      struct vhost_virtqueue *vq)
 {
+       struct vhost_net_virtqueue *nvq =
+               container_of(vq, struct vhost_net_virtqueue, vq);
        int i;
        int j = 0;
 
-       for (i = vq->done_idx; i != vq->upend_idx; i = (i + 1) % UIO_MAXIOV) {
+       for (i = nvq->done_idx; i != nvq->upend_idx; i = (i + 1) % UIO_MAXIOV) {
                if (vq->heads[i].len == VHOST_DMA_FAILED_LEN)
                        vhost_net_tx_err(net);
                if (VHOST_DMA_IS_DONE(vq->heads[i].len)) {
@@ -172,7 +272,7 @@ static int vhost_zerocopy_signal_used(struct vhost_net *net,
                        break;
        }
        if (j)
-               vq->done_idx = i;
+               nvq->done_idx = i;
        return j;
 }
 
@@ -203,6 +303,7 @@ static void vhost_zerocopy_callback(struct ubuf_info *ubuf, bool success)
 static void handle_tx(struct vhost_net *net)
 {
        struct vhost_virtqueue *vq = &net->vqs[VHOST_NET_VQ_TX].vq;
+       struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX];
        unsigned out, in, s;
        int head;
        struct msghdr msg = {
@@ -229,7 +330,7 @@ static void handle_tx(struct vhost_net *net)
        vhost_disable_notify(&net->dev, vq);
 
        hdr_size = vq->vhost_hlen;
-       zcopy = vq->ubufs;
+       zcopy = nvq->ubufs;
 
        for (;;) {
                /* Release DMAs done buffers first */
@@ -250,9 +351,10 @@ static void handle_tx(struct vhost_net *net)
                        /* If more outstanding DMAs, queue the work.
                         * Handle upend_idx wrap around
                         */
-                       num_pends = likely(vq->upend_idx >= vq->done_idx) ?
-                                   (vq->upend_idx - vq->done_idx) :
-                                   (vq->upend_idx + UIO_MAXIOV - vq->done_idx);
+                       num_pends = likely(nvq->upend_idx >= nvq->done_idx) ?
+                                   (nvq->upend_idx - nvq->done_idx) :
+                                   (nvq->upend_idx + UIO_MAXIOV -
+                                    nvq->done_idx);
                        if (unlikely(num_pends > VHOST_MAX_PEND))
                                break;
                        if (unlikely(vhost_enable_notify(&net->dev, vq))) {
@@ -278,34 +380,34 @@ static void handle_tx(struct vhost_net *net)
                        break;
                }
                zcopy_used = zcopy && (len >= VHOST_GOODCOPY_LEN ||
-                                      vq->upend_idx != vq->done_idx);
+                                      nvq->upend_idx != nvq->done_idx);
 
                /* use msg_control to pass vhost zerocopy ubuf info to skb */
                if (zcopy_used) {
-                       vq->heads[vq->upend_idx].id = head;
+                       vq->heads[nvq->upend_idx].id = head;
                        if (!vhost_net_tx_select_zcopy(net) ||
                            len < VHOST_GOODCOPY_LEN) {
                                /* copy don't need to wait for DMA done */
-                               vq->heads[vq->upend_idx].len =
+                               vq->heads[nvq->upend_idx].len =
                                                        VHOST_DMA_DONE_LEN;
                                msg.msg_control = NULL;
                                msg.msg_controllen = 0;
                                ubufs = NULL;
                        } else {
                                struct ubuf_info *ubuf;
-                               ubuf = vq->ubuf_info + vq->upend_idx;
+                               ubuf = nvq->ubuf_info + nvq->upend_idx;
 
-                               vq->heads[vq->upend_idx].len =
+                               vq->heads[nvq->upend_idx].len =
                                        VHOST_DMA_IN_PROGRESS;
                                ubuf->callback = vhost_zerocopy_callback;
-                               ubuf->ctx = vq->ubufs;
-                               ubuf->desc = vq->upend_idx;
+                               ubuf->ctx = nvq->ubufs;
+                               ubuf->desc = nvq->upend_idx;
                                msg.msg_control = ubuf;
                                msg.msg_controllen = sizeof(ubuf);
-                               ubufs = vq->ubufs;
+                               ubufs = nvq->ubufs;
                                kref_get(&ubufs->kref);
                        }
-                       vq->upend_idx = (vq->upend_idx + 1) % UIO_MAXIOV;
+                       nvq->upend_idx = (nvq->upend_idx + 1) % UIO_MAXIOV;
                }
                /* TODO: Check specific error and bomb out unless ENOBUFS? */
                err = sock->ops->sendmsg(NULL, sock, &msg, len);
@@ -313,8 +415,8 @@ static void handle_tx(struct vhost_net *net)
                        if (zcopy_used) {
                                if (ubufs)
                                        vhost_ubuf_put(ubufs);
-                               vq->upend_idx = ((unsigned)vq->upend_idx - 1) %
-                                       UIO_MAXIOV;
+                               nvq->upend_idx = ((unsigned)nvq->upend_idx - 1)
+                                       UIO_MAXIOV;
                        }
                        vhost_discard_vq_desc(vq, 1);
                        break;
@@ -564,7 +666,7 @@ static int vhost_net_open(struct inode *inode, struct file *f)
        struct vhost_net *n = kmalloc(sizeof *n, GFP_KERNEL);
        struct vhost_dev *dev;
        struct vhost_virtqueue **vqs;
-       int r;
+       int r, i;
 
        if (!n)
                return -ENOMEM;
@@ -579,6 +681,12 @@ static int vhost_net_open(struct inode *inode, struct file *f)
        vqs[VHOST_NET_VQ_RX] = &n->vqs[VHOST_NET_VQ_RX].vq;
        n->vqs[VHOST_NET_VQ_TX].vq.handle_kick = handle_tx_kick;
        n->vqs[VHOST_NET_VQ_RX].vq.handle_kick = handle_rx_kick;
+       for (i = 0; i < VHOST_NET_VQ_MAX; i++) {
+               n->vqs[i].ubufs = NULL;
+               n->vqs[i].ubuf_info = NULL;
+               n->vqs[i].upend_idx = 0;
+               n->vqs[i].done_idx = 0;
+       }
        r = vhost_dev_init(dev, vqs, VHOST_NET_VQ_MAX);
        if (r < 0) {
                kfree(n);
@@ -652,15 +760,15 @@ static void vhost_net_flush(struct vhost_net *n)
 {
        vhost_net_flush_vq(n, VHOST_NET_VQ_TX);
        vhost_net_flush_vq(n, VHOST_NET_VQ_RX);
-       if (n->vqs[VHOST_NET_VQ_TX].vq.ubufs) {
+       if (n->vqs[VHOST_NET_VQ_TX].ubufs) {
                mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
                n->tx_flush = true;
                mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
                /* Wait for all lower device DMAs done. */
-               vhost_ubuf_put_and_wait(n->vqs[VHOST_NET_VQ_TX].vq.ubufs);
+               vhost_ubuf_put_and_wait(n->vqs[VHOST_NET_VQ_TX].ubufs);
                mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
                n->tx_flush = false;
-               kref_init(&n->vqs[VHOST_NET_VQ_TX].vq.ubufs->kref);
+               kref_init(&n->vqs[VHOST_NET_VQ_TX].ubufs->kref);
                mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex);
        }
 }
@@ -675,6 +783,7 @@ static int vhost_net_release(struct inode *inode, struct file *f)
        vhost_net_flush(n);
        vhost_dev_stop(&n->dev);
        vhost_dev_cleanup(&n->dev, false);
+       vhost_net_reset_ubuf_info(n);
        if (tx_sock)
                fput(tx_sock->file);
        if (rx_sock)
@@ -756,6 +865,7 @@ static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
 {
        struct socket *sock, *oldsock;
        struct vhost_virtqueue *vq;
+       struct vhost_net_virtqueue *nvq;
        struct vhost_ubuf_ref *ubufs, *oldubufs = NULL;
        int r;
 
@@ -769,6 +879,7 @@ static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
                goto err;
        }
        vq = &n->vqs[index].vq;
+       nvq = &n->vqs[index];
        mutex_lock(&vq->mutex);
 
        /* Verify that ring has been setup correctly. */
@@ -801,8 +912,8 @@ static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
                if (r)
                        goto err_used;
 
-               oldubufs = vq->ubufs;
-               vq->ubufs = ubufs;
+               oldubufs = nvq->ubufs;
+               nvq->ubufs = ubufs;
 
                n->tx_packets = 0;
                n->tx_zcopy_err = 0;
@@ -853,6 +964,7 @@ static long vhost_net_reset_owner(struct vhost_net *n)
        vhost_net_stop(n, &tx_sock, &rx_sock);
        vhost_net_flush(n);
        err = vhost_dev_reset_owner(&n->dev);
+       vhost_net_reset_ubuf_info(n);
 done:
        mutex_unlock(&n->dev.mutex);
        if (tx_sock)
@@ -928,11 +1040,17 @@ static long vhost_net_ioctl(struct file *f, unsigned int ioctl,
                return vhost_net_reset_owner(n);
        default:
                mutex_lock(&n->dev.mutex);
+               if (ioctl == VHOST_SET_OWNER) {
+                       r = vhost_net_set_ubuf_info(n);
+                       if (r)
+                               goto out;
+               }
                r = vhost_dev_ioctl(&n->dev, ioctl, argp);
                if (r == -ENOIOCTLCMD)
                        r = vhost_vring_ioctl(&n->dev, ioctl, argp);
                else
                        vhost_net_flush(n);
+out:
                mutex_unlock(&n->dev.mutex);
                return r;
        }
index bef8b6bae186d4b5c1b544139a3c4dc446dca3e9..6644812e99b4e8fccaff20664ba6ce8ffa56e799 100644 (file)
@@ -33,8 +33,6 @@ enum {
        VHOST_MEMORY_F_LOG = 0x1,
 };
 
-static unsigned vhost_zcopy_mask __read_mostly;
-
 #define vhost_used_event(vq) ((u16 __user *)&vq->avail->ring[vq->num])
 #define vhost_avail_event(vq) ((u16 __user *)&vq->used->ring[vq->num])
 
@@ -191,9 +189,6 @@ static void vhost_vq_reset(struct vhost_dev *dev,
        vq->call_ctx = NULL;
        vq->call = NULL;
        vq->log_ctx = NULL;
-       vq->upend_idx = 0;
-       vq->done_idx = 0;
-       vq->ubufs = NULL;
 }
 
 static int vhost_worker(void *data)
@@ -253,20 +248,12 @@ static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq)
        vq->log = NULL;
        kfree(vq->heads);
        vq->heads = NULL;
-       kfree(vq->ubuf_info);
-       vq->ubuf_info = NULL;
-}
-
-void vhost_enable_zcopy(int vq)
-{
-       vhost_zcopy_mask |= 0x1 << vq;
 }
 
 /* Helper to allocate iovec buffers for all vqs. */
 static long vhost_dev_alloc_iovecs(struct vhost_dev *dev)
 {
        int i;
-       bool zcopy;
 
        for (i = 0; i < dev->nvqs; ++i) {
                dev->vqs[i]->indirect = kmalloc(sizeof *dev->vqs[i]->indirect *
@@ -275,14 +262,8 @@ static long vhost_dev_alloc_iovecs(struct vhost_dev *dev)
                                          GFP_KERNEL);
                dev->vqs[i]->heads = kmalloc(sizeof *dev->vqs[i]->heads *
                                            UIO_MAXIOV, GFP_KERNEL);
-               zcopy = vhost_zcopy_mask & (0x1 << i);
-               if (zcopy)
-                       dev->vqs[i]->ubuf_info =
-                               kmalloc(sizeof *dev->vqs[i]->ubuf_info *
-                                       UIO_MAXIOV, GFP_KERNEL);
                if (!dev->vqs[i]->indirect || !dev->vqs[i]->log ||
-                       !dev->vqs[i]->heads ||
-                       (zcopy && !dev->vqs[i]->ubuf_info))
+                       !dev->vqs[i]->heads)
                        goto err_nomem;
        }
        return 0;
@@ -321,7 +302,6 @@ long vhost_dev_init(struct vhost_dev *dev,
                dev->vqs[i]->log = NULL;
                dev->vqs[i]->indirect = NULL;
                dev->vqs[i]->heads = NULL;
-               dev->vqs[i]->ubuf_info = NULL;
                dev->vqs[i]->dev = dev;
                mutex_init(&dev->vqs[i]->mutex);
                vhost_vq_reset(dev, dev->vqs[i]);
@@ -1551,38 +1531,3 @@ void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq)
                               &vq->used->flags, r);
        }
 }
-
-static void vhost_zerocopy_done_signal(struct kref *kref)
-{
-       struct vhost_ubuf_ref *ubufs = container_of(kref, struct vhost_ubuf_ref,
-                                                   kref);
-       wake_up(&ubufs->wait);
-}
-
-struct vhost_ubuf_ref *vhost_ubuf_alloc(struct vhost_virtqueue *vq,
-                                       bool zcopy)
-{
-       struct vhost_ubuf_ref *ubufs;
-       /* No zero copy backend? Nothing to count. */
-       if (!zcopy)
-               return NULL;
-       ubufs = kmalloc(sizeof *ubufs, GFP_KERNEL);
-       if (!ubufs)
-               return ERR_PTR(-ENOMEM);
-       kref_init(&ubufs->kref);
-       init_waitqueue_head(&ubufs->wait);
-       ubufs->vq = vq;
-       return ubufs;
-}
-
-void vhost_ubuf_put(struct vhost_ubuf_ref *ubufs)
-{
-       kref_put(&ubufs->kref, vhost_zerocopy_done_signal);
-}
-
-void vhost_ubuf_put_and_wait(struct vhost_ubuf_ref *ubufs)
-{
-       kref_put(&ubufs->kref, vhost_zerocopy_done_signal);
-       wait_event(ubufs->wait, !atomic_read(&ubufs->kref.refcount));
-       kfree(ubufs);
-}
index f3afa8a41fe0fb2dc29eb2a2bc5ea60ccdbc59b3..3a36712e0792b33eed87766fcf0344ddbb4564be 100644 (file)
@@ -54,18 +54,6 @@ struct vhost_log {
 
 struct vhost_virtqueue;
 
-struct vhost_ubuf_ref {
-       struct kref kref;
-       wait_queue_head_t wait;
-       struct vhost_virtqueue *vq;
-};
-
-struct vhost_ubuf_ref *vhost_ubuf_alloc(struct vhost_virtqueue *, bool zcopy);
-void vhost_ubuf_put(struct vhost_ubuf_ref *);
-void vhost_ubuf_put_and_wait(struct vhost_ubuf_ref *);
-
-struct ubuf_info;
-
 /* The virtqueue structure describes a queue attached to a device. */
 struct vhost_virtqueue {
        struct vhost_dev *dev;
@@ -130,16 +118,6 @@ struct vhost_virtqueue {
        /* Log write descriptors */
        void __user *log_base;
        struct vhost_log *log;
-       /* vhost zerocopy support fields below: */
-       /* last used idx for outstanding DMA zerocopy buffers */
-       int upend_idx;
-       /* first used idx for DMA done zerocopy buffers */
-       int done_idx;
-       /* an array of userspace buffers info */
-       struct ubuf_info *ubuf_info;
-       /* Reference counting for outstanding ubufs.
-        * Protected by vq mutex. Writers must also take device mutex. */
-       struct vhost_ubuf_ref *ubufs;
 };
 
 struct vhost_dev {