@@ -32,6 +32,10 @@
* Using this limit prevents one virtqueue from starving others. */
#define VHOST_NET_WEIGHT 0x80000
+/* MAX number of TX used buffers for outstanding zerocopy */
+#define VHOST_MAX_PEND 128
+#define VHOST_GOODCOPY_LEN 256
+
enum {
VHOST_NET_VQ_RX = 0,
VHOST_NET_VQ_TX = 1,
@@ -129,6 +133,7 @@ static void handle_tx(struct vhost_net *net)
int err, wmem;
size_t hdr_size;
struct socket *sock;
+ struct skb_ubuf_info pend;
/* TODO: check that we are running from vhost_worker? */
sock = rcu_dereference_check(vq->private_data, 1);
@@ -151,6 +156,10 @@ static void handle_tx(struct vhost_net *net)
hdr_size = vq->vhost_hlen;
for (;;) {
+ /* Release DMAs done buffers first */
+ if (atomic_read(&vq->refcnt) > VHOST_MAX_PEND)
+ vhost_zerocopy_signal_used(vq, false);
+
head = vhost_get_vq_desc(&net->dev, vq, vq->iov,
ARRAY_SIZE(vq->iov),
&out, &in,
@@ -166,6 +175,12 @@ static void handle_tx(struct vhost_net *net)
set_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
break;
}
+ /* If more outstanding DMAs, queue the work */
+ if (atomic_read(&vq->refcnt) > VHOST_MAX_PEND) {
+ tx_poll_start(net, sock);
+ set_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
+ break;
+ }
if (unlikely(vhost_enable_notify(vq))) {
vhost_disable_notify(vq);
continue;
@@ -188,6 +203,24 @@ static void handle_tx(struct vhost_net *net)
iov_length(vq->hdr, s), hdr_size);
break;
}
+ /* use msg_control to pass vhost zerocopy ubuf info to skb */
+ if (sock_flag(sock->sk, SOCK_ZEROCOPY)) {
+ vq->heads[vq->upend_idx].id = head;
+ if (len < VHOST_GOODCOPY_LEN)
+ /* copy don't need to wait for DMA done */
+ vq->heads[vq->upend_idx].len =
+ VHOST_DMA_DONE_LEN;
+ else {
+ vq->heads[vq->upend_idx].len = len;
+ pend.callback = vhost_zerocopy_callback;
+ pend.arg = vq;
+ pend.desc = vq->upend_idx;
+ msg.msg_control = &pend;
+ msg.msg_controllen = sizeof(pend);
+ }
+ atomic_inc(&vq->refcnt);
+ vq->upend_idx = (vq->upend_idx + 1) % UIO_MAXIOV;
+ }
/* TODO: Check specific error and bomb out unless ENOBUFS? */
err = sock->ops->sendmsg(NULL, sock, &msg, len);
if (unlikely(err < 0)) {
@@ -198,12 +231,21 @@ static void handle_tx(struct vhost_net *net)
if (err != len)
pr_debug("Truncated TX packet: "
" len %d != %zd\n", err, len);
- vhost_add_used_and_signal(&net->dev, vq, head, 0);
+ if (!sock_flag(sock->sk, SOCK_ZEROCOPY))
+ vhost_add_used_and_signal(&net->dev, vq, head, 0);
total_len += len;
if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
vhost_poll_queue(&vq->poll);
break;
}
+ /* if upend_idx is full, then wait for free more */
+/*
+ if (unlikely(vq->upend_idx == vq->done_idx)) {
+ tx_poll_start(net, sock);
+ set_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
+ break;
+ }
+*/
}
mutex_unlock(&vq->mutex);
@@ -174,6 +174,9 @@ 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;
+ atomic_set(&vq->refcnt, 0);
}
static int vhost_worker(void *data)
@@ -385,16 +388,49 @@ long vhost_dev_reset_owner(struct vhost_dev *dev)
return 0;
}
+/* In case of DMA done not in order in lower device driver for some reason.
+ * upend_idx is used to track end of used idx, done_idx is used to track head
+ * of used idx. Once lower device DMA done contiguously, we will signal KVM
+ * guest used idx.
+ */
+void vhost_zerocopy_signal_used(struct vhost_virtqueue *vq, bool shutdown)
+{
+ int i, j = 0;
+
+ for (i = vq->done_idx; i != vq->upend_idx; i = ++i % UIO_MAXIOV) {
+ if ((vq->heads[i].len == VHOST_DMA_DONE_LEN) || shutdown) {
+ vq->heads[i].len = VHOST_DMA_CLEAR_LEN;
+ vhost_add_used_and_signal(vq->dev, vq,
+ vq->heads[i].id, 0);
+ ++j;
+ } else
+ break;
+ }
+ if (j) {
+ vq->done_idx = i;
+ atomic_sub(j, &vq->refcnt);
+ }
+}
+
/* Caller should have device mutex */
void vhost_dev_cleanup(struct vhost_dev *dev)
{
int i;
+ unsigned long begin = jiffies;
for (i = 0; i < dev->nvqs; ++i) {
if (dev->vqs[i].kick && dev->vqs[i].handle_kick) {
vhost_poll_stop(&dev->vqs[i].poll);
vhost_poll_flush(&dev->vqs[i].poll);
}
+ /* Wait for all lower device DMAs done, then notify virtio_net
+ * or just notify it without waiting for all DMA done here ?
+ * in case of DMAs never done for some reason */
+ if (atomic_read(&dev->vqs[i].refcnt)) {
+ /* how long should we wait ? */
+ msleep(1000);
+ vhost_zerocopy_signal_used(&dev->vqs[i], true);
+ }
if (dev->vqs[i].error_ctx)
eventfd_ctx_put(dev->vqs[i].error_ctx);
if (dev->vqs[i].error)
@@ -603,6 +639,10 @@ static long vhost_set_vring(struct vhost_dev *d, int ioctl, void __user *argp)
mutex_lock(&vq->mutex);
+ /* clean up lower device outstanding DMAs, before setting ring */
+ if (atomic_read(&vq->refcnt))
+ vhost_zerocopy_signal_used(vq, true);
+
switch (ioctl) {
case VHOST_SET_VRING_NUM:
/* Resizing ring with an active backend?
@@ -1416,3 +1456,12 @@ void vhost_disable_notify(struct vhost_virtqueue *vq)
vq_err(vq, "Failed to enable notification at %p: %d\n",
&vq->used->flags, r);
}
+
+void vhost_zerocopy_callback(struct sk_buff *skb)
+{
+ int idx = skb_shinfo(skb)->ubuf.desc;
+ struct vhost_virtqueue *vq = skb_shinfo(skb)->ubuf.arg;
+
+ /* set len = 1 to mark this desc buffers done DMA */
+ vq->heads[idx].len = VHOST_DMA_DONE_LEN;
+}
@@ -13,6 +13,11 @@
#include <linux/virtio_ring.h>
#include <asm/atomic.h>
+/* This is for zerocopy, used buffer len is set to 1 when lower device DMA
+ * done */
+#define VHOST_DMA_DONE_LEN 1
+#define VHOST_DMA_CLEAR_LEN 0
+
struct vhost_device;
struct vhost_work;
@@ -108,6 +113,12 @@ struct vhost_virtqueue {
/* Log write descriptors */
void __user *log_base;
struct vhost_log *log;
+ /* vhost zerocopy support */
+ atomic_t refcnt; /* num of outstanding zerocopy DMAs */
+ /* last used idx for outstanding DMA zerocopy buffers */
+ int upend_idx;
+ /* first used idx for DMA done zerocopy buffers */
+ int done_idx;
};
struct vhost_dev {
@@ -154,6 +165,8 @@ bool vhost_enable_notify(struct vhost_virtqueue *);
int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
unsigned int log_num, u64 len);
+void vhost_zerocopy_callback(struct sk_buff *skb);
+void vhost_zerocopy_signal_used(struct vhost_virtqueue *vq, bool shutdown);
#define vq_err(vq, fmt, ...) do { \
pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \
Hello Michael, Let me anything I might miss from your preview's review. Thanks Shirley ----------- This patch maintains the outstanding userspace buffers in the sequence it is delivered to vhost. The outstanding userspace buffers will be marked as done once the lower device buffers DMA has finished. This is monitored through last reference of kfree_skb callback. Two buffer index are used for this purpose. The vhost passes the userspace buffers info to lower device skb through message control. Since there will be some done DMAs when entering vhost handle_tx. The worse case is all buffers in the vq are in pending/done status, so we need to notify guest to release DMA done buffers first before get any new buffers from the vq. Signed-off-by: Shirley <xma@us.ibm.com> --- drivers/vhost/net.c | 44 +++++++++++++++++++++++++++++++++++++++++++- drivers/vhost/vhost.c | 49 +++++++++++++++++++++++++++++++++++++++++++++++++ drivers/vhost/vhost.h | 13 +++++++++++++ 3 files changed, 105 insertions(+), 1 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html