[1/6] vhost: allow device that does not depend on vhost worker
diff mbox series

Message ID 20200529080303.15449-2-jasowang@redhat.com
State New
Headers show
Series
  • vDPA: doorbell mapping
Related show

Commit Message

Jason Wang May 29, 2020, 8:02 a.m. UTC
vDPA device currently relays the eventfd via vhost worker. This is
inefficient due the latency of wakeup and scheduling, so this patch
tries to introduce a use_worker attribute for the vhost device. When
use_worker is not set with vhost_dev_init(), vhost won't try to
allocate a worker thread and the vhost_poll will be processed directly
in the wakeup function.

This help for vDPA since it reduces the latency caused by vhost worker.

In my testing, it saves 0.2 ms in pings between VMs on a mutual host.

Signed-off-by: Zhu Lingshan <lingshan.zhu@intel.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
---
 drivers/vhost/net.c   |  2 +-
 drivers/vhost/scsi.c  |  2 +-
 drivers/vhost/vdpa.c  |  2 +-
 drivers/vhost/vhost.c | 38 +++++++++++++++++++++++++-------------
 drivers/vhost/vhost.h |  2 ++
 drivers/vhost/vsock.c |  2 +-
 6 files changed, 31 insertions(+), 17 deletions(-)

Comments

Michael S. Tsirkin June 2, 2020, 5:01 a.m. UTC | #1
On Fri, May 29, 2020 at 04:02:58PM +0800, Jason Wang wrote:
> diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
> index d450e16c5c25..70105e045768 100644
> --- a/drivers/vhost/vhost.c
> +++ b/drivers/vhost/vhost.c
> @@ -166,11 +166,16 @@ static int vhost_poll_wakeup(wait_queue_entry_t *wait, unsigned mode, int sync,
>  			     void *key)
>  {
>  	struct vhost_poll *poll = container_of(wait, struct vhost_poll, wait);
> +	struct vhost_work *work = &poll->work;
>  
>  	if (!(key_to_poll(key) & poll->mask))
>  		return 0;
>  
> -	vhost_poll_queue(poll);
> +	if (!poll->dev->use_worker)
> +		work->fn(work);
> +	else
> +		vhost_poll_queue(poll);
> +
>  	return 0;
>  }
>

So a wakeup function wakes up eventfd directly.

What if user supplies e.g. the same eventfd as ioeventfd?

Won't this cause infinite loops?
Jason Wang June 2, 2020, 7:04 a.m. UTC | #2
On 2020/6/2 下午1:01, Michael S. Tsirkin wrote:
> On Fri, May 29, 2020 at 04:02:58PM +0800, Jason Wang wrote:
>> diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
>> index d450e16c5c25..70105e045768 100644
>> --- a/drivers/vhost/vhost.c
>> +++ b/drivers/vhost/vhost.c
>> @@ -166,11 +166,16 @@ static int vhost_poll_wakeup(wait_queue_entry_t *wait, unsigned mode, int sync,
>>   			     void *key)
>>   {
>>   	struct vhost_poll *poll = container_of(wait, struct vhost_poll, wait);
>> +	struct vhost_work *work = &poll->work;
>>   
>>   	if (!(key_to_poll(key) & poll->mask))
>>   		return 0;
>>   
>> -	vhost_poll_queue(poll);
>> +	if (!poll->dev->use_worker)
>> +		work->fn(work);
>> +	else
>> +		vhost_poll_queue(poll);
>> +
>>   	return 0;
>>   }
>>
> So a wakeup function wakes up eventfd directly.
>
> What if user supplies e.g. the same eventfd as ioeventfd?
>
> Won't this cause infinite loops?


I'm not sure I get this.

This basically calls handle_vq directly when eventfd is woken up. The 
infinite loops can only happen when handle_vq() tries to write to 
ioeventfd itslef which should be a bug of the device.

Thanks


>

Patch
diff mbox series

diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
index 2927f02cc7e1..bf5e1d81ae25 100644
--- a/drivers/vhost/net.c
+++ b/drivers/vhost/net.c
@@ -1326,7 +1326,7 @@  static int vhost_net_open(struct inode *inode, struct file *f)
 	}
 	vhost_dev_init(dev, vqs, VHOST_NET_VQ_MAX,
 		       UIO_MAXIOV + VHOST_NET_BATCH,
-		       VHOST_NET_PKT_WEIGHT, VHOST_NET_WEIGHT,
+		       VHOST_NET_PKT_WEIGHT, VHOST_NET_WEIGHT, true,
 		       NULL);
 
 	vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, EPOLLOUT, dev);
diff --git a/drivers/vhost/scsi.c b/drivers/vhost/scsi.c
index c39952243fd3..0cbaa0b3893d 100644
--- a/drivers/vhost/scsi.c
+++ b/drivers/vhost/scsi.c
@@ -1628,7 +1628,7 @@  static int vhost_scsi_open(struct inode *inode, struct file *f)
 		vs->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
 	}
 	vhost_dev_init(&vs->dev, vqs, VHOST_SCSI_MAX_VQ, UIO_MAXIOV,
-		       VHOST_SCSI_WEIGHT, 0, NULL);
+		       VHOST_SCSI_WEIGHT, 0, true, NULL);
 
 	vhost_scsi_init_inflight(vs, NULL);
 
diff --git a/drivers/vhost/vdpa.c b/drivers/vhost/vdpa.c
index 20476b505d99..6ff72289f488 100644
--- a/drivers/vhost/vdpa.c
+++ b/drivers/vhost/vdpa.c
@@ -696,7 +696,7 @@  static int vhost_vdpa_open(struct inode *inode, struct file *filep)
 		vqs[i] = &v->vqs[i];
 		vqs[i]->handle_kick = handle_vq_kick;
 	}
-	vhost_dev_init(dev, vqs, nvqs, 0, 0, 0,
+	vhost_dev_init(dev, vqs, nvqs, 0, 0, 0, false,
 		       vhost_vdpa_process_iotlb_msg);
 
 	dev->iotlb = vhost_iotlb_alloc(0, 0);
diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index d450e16c5c25..70105e045768 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -166,11 +166,16 @@  static int vhost_poll_wakeup(wait_queue_entry_t *wait, unsigned mode, int sync,
 			     void *key)
 {
 	struct vhost_poll *poll = container_of(wait, struct vhost_poll, wait);
+	struct vhost_work *work = &poll->work;
 
 	if (!(key_to_poll(key) & poll->mask))
 		return 0;
 
-	vhost_poll_queue(poll);
+	if (!poll->dev->use_worker)
+		work->fn(work);
+	else
+		vhost_poll_queue(poll);
+
 	return 0;
 }
 
@@ -454,6 +459,7 @@  static size_t vhost_get_desc_size(struct vhost_virtqueue *vq,
 void vhost_dev_init(struct vhost_dev *dev,
 		    struct vhost_virtqueue **vqs, int nvqs,
 		    int iov_limit, int weight, int byte_weight,
+		    bool use_worker,
 		    int (*msg_handler)(struct vhost_dev *dev,
 				       struct vhost_iotlb_msg *msg))
 {
@@ -471,6 +477,7 @@  void vhost_dev_init(struct vhost_dev *dev,
 	dev->iov_limit = iov_limit;
 	dev->weight = weight;
 	dev->byte_weight = byte_weight;
+	dev->use_worker = use_worker;
 	dev->msg_handler = msg_handler;
 	init_llist_head(&dev->work_list);
 	init_waitqueue_head(&dev->wait);
@@ -549,18 +556,21 @@  long vhost_dev_set_owner(struct vhost_dev *dev)
 	/* No owner, become one */
 	dev->mm = get_task_mm(current);
 	dev->kcov_handle = kcov_common_handle();
-	worker = kthread_create(vhost_worker, dev, "vhost-%d", current->pid);
-	if (IS_ERR(worker)) {
-		err = PTR_ERR(worker);
-		goto err_worker;
-	}
+	if (dev->use_worker) {
+		worker = kthread_create(vhost_worker, dev,
+					"vhost-%d", current->pid);
+		if (IS_ERR(worker)) {
+			err = PTR_ERR(worker);
+			goto err_worker;
+		}
 
-	dev->worker = worker;
-	wake_up_process(worker);	/* avoid contributing to loadavg */
+		dev->worker = worker;
+		wake_up_process(worker); /* avoid contributing to loadavg */
 
-	err = vhost_attach_cgroups(dev);
-	if (err)
-		goto err_cgroup;
+		err = vhost_attach_cgroups(dev);
+		if (err)
+			goto err_cgroup;
+	}
 
 	err = vhost_dev_alloc_iovecs(dev);
 	if (err)
@@ -568,8 +578,10 @@  long vhost_dev_set_owner(struct vhost_dev *dev)
 
 	return 0;
 err_cgroup:
-	kthread_stop(worker);
-	dev->worker = NULL;
+	if (dev->worker) {
+		kthread_stop(dev->worker);
+		dev->worker = NULL;
+	}
 err_worker:
 	if (dev->mm)
 		mmput(dev->mm);
diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h
index f8403bd46b85..0feb6701e273 100644
--- a/drivers/vhost/vhost.h
+++ b/drivers/vhost/vhost.h
@@ -154,6 +154,7 @@  struct vhost_dev {
 	int weight;
 	int byte_weight;
 	u64 kcov_handle;
+	bool use_worker;
 	int (*msg_handler)(struct vhost_dev *dev,
 			   struct vhost_iotlb_msg *msg);
 };
@@ -161,6 +162,7 @@  struct vhost_dev {
 bool vhost_exceeds_weight(struct vhost_virtqueue *vq, int pkts, int total_len);
 void vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue **vqs,
 		    int nvqs, int iov_limit, int weight, int byte_weight,
+		    bool use_worker,
 		    int (*msg_handler)(struct vhost_dev *dev,
 				       struct vhost_iotlb_msg *msg));
 long vhost_dev_set_owner(struct vhost_dev *dev);
diff --git a/drivers/vhost/vsock.c b/drivers/vhost/vsock.c
index e36aaf9ba7bd..2eb85c42bac4 100644
--- a/drivers/vhost/vsock.c
+++ b/drivers/vhost/vsock.c
@@ -621,7 +621,7 @@  static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
 
 	vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
 		       UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT,
-		       VHOST_VSOCK_WEIGHT, NULL);
+		       VHOST_VSOCK_WEIGHT, true, NULL);
 
 	file->private_data = vsock;
 	spin_lock_init(&vsock->send_pkt_list_lock);