[PATCH v1 0/2] virtio-balloon: tweak config_changed

[PATCH v1 0/2] virtio-balloon: tweak config_changed

[Wei Wang]

Since virtio-ccw doesn't work with accessing to the config registers
inside an interrupt context, this patch series avoids that issue by
moving the config register accesses to the related workqueue contexts.

Wei Wang (2):
  virtio-balloon: tweak config_changed implementation
  virtio-balloon: improve update_balloon_size_func

 drivers/virtio/virtio_balloon.c | 59 +++++++++++++++++++++--------------------
 1 file changed, 30 insertions(+), 29 deletions(-)

-- 
2.7.4

[PATCH v1 1/2] virtio-balloon: tweak config_changed implementation

[Wei Wang]

virtio-ccw has deadlock issues with reading config registers inside the
interrupt context, so we tweak the virtballoon_changed implementation
by moving the config read operations into the related workqueue contexts.

Signed-off-by: Wei Wang <wei.w.wang@intel.com>
---
 drivers/virtio/virtio_balloon.c | 54 ++++++++++++++++++++---------------------
 1 file changed, 26 insertions(+), 28 deletions(-)

diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
index 728ecd1..9a82a11 100644
--- a/drivers/virtio/virtio_balloon.c
+++ b/drivers/virtio/virtio_balloon.c
@@ -394,33 +394,15 @@ static void virtballoon_changed(struct virtio_device *vdev)
 {
 	struct virtio_balloon *vb = vdev->priv;
 	unsigned long flags;
-	s64 diff = towards_target(vb);
-
-	if (diff) {
-		spin_lock_irqsave(&vb->stop_update_lock, flags);
-		if (!vb->stop_update)
-			queue_work(system_freezable_wq,
-				   &vb->update_balloon_size_work);
-		spin_unlock_irqrestore(&vb->stop_update_lock, flags);
-	}
 
-	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
-		virtio_cread(vdev, struct virtio_balloon_config,
-			     free_page_report_cmd_id, &vb->cmd_id_received);
-		if (vb->cmd_id_received == VIRTIO_BALLOON_CMD_ID_DONE) {
-			/* Pass ULONG_MAX to give back all the free pages */
-			return_free_pages_to_mm(vb, ULONG_MAX);
-		} else if (vb->cmd_id_received != VIRTIO_BALLOON_CMD_ID_STOP &&
-			   vb->cmd_id_received !=
-			   virtio32_to_cpu(vdev, vb->cmd_id_active)) {
-			spin_lock_irqsave(&vb->stop_update_lock, flags);
-			if (!vb->stop_update) {
-				queue_work(vb->balloon_wq,
-					   &vb->report_free_page_work);
-			}
-			spin_unlock_irqrestore(&vb->stop_update_lock, flags);
-		}
+	spin_lock_irqsave(&vb->stop_update_lock, flags);
+	if (!vb->stop_update) {
+		queue_work(system_freezable_wq,
+			   &vb->update_balloon_size_work);
+		if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
+			queue_work(vb->balloon_wq, &vb->report_free_page_work);
 	}
+	spin_unlock_irqrestore(&vb->stop_update_lock, flags);
 }
 
 static void update_balloon_size(struct virtio_balloon *vb)
@@ -637,11 +619,9 @@ static int send_free_pages(struct virtio_balloon *vb)
 	return 0;
 }
 
-static void report_free_page_func(struct work_struct *work)
+static void virtio_balloon_report_free_page(struct virtio_balloon *vb)
 {
 	int err;
-	struct virtio_balloon *vb = container_of(work, struct virtio_balloon,
-						 report_free_page_work);
 	struct device *dev = &vb->vdev->dev;
 
 	/* Start by sending the received cmd id to host with an outbuf. */
@@ -659,6 +639,24 @@ static void report_free_page_func(struct work_struct *work)
 		dev_err(dev, "Failed to send a stop id, err = %d\n", err);
 }
 
+static void report_free_page_func(struct work_struct *work)
+{
+	struct virtio_balloon *vb = container_of(work, struct virtio_balloon,
+						 report_free_page_work);
+
+	virtio_cread(vb->vdev, struct virtio_balloon_config,
+		     free_page_report_cmd_id, &vb->cmd_id_received);
+
+	if (vb->cmd_id_received == VIRTIO_BALLOON_CMD_ID_DONE) {
+		/* Pass ULONG_MAX to give back all the free pages */
+		return_free_pages_to_mm(vb, ULONG_MAX);
+	} else if (vb->cmd_id_received != VIRTIO_BALLOON_CMD_ID_STOP &&
+		   vb->cmd_id_received !=
+		   virtio32_to_cpu(vb->vdev, vb->cmd_id_active)) {
+		virtio_balloon_report_free_page(vb);
+	}
+}
+
 #ifdef CONFIG_BALLOON_COMPACTION
 /*
  * virtballoon_migratepage - perform the balloon page migration on behalf of
-- 
2.7.4

[PATCH v1 2/2] virtio-balloon: improve update_balloon_size_func

[Wei Wang]

There is no need to update the balloon actual register when there is no
ballooning request. This patch avoids update_balloon_size when diff is 0.

Signed-off-by: Wei Wang <wei.w.wang@intel.com>
---
 drivers/virtio/virtio_balloon.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
index 9a82a11..4884b85 100644
--- a/drivers/virtio/virtio_balloon.c
+++ b/drivers/virtio/virtio_balloon.c
@@ -435,9 +435,12 @@ static void update_balloon_size_func(struct work_struct *work)
 			  update_balloon_size_work);
 	diff = towards_target(vb);
 
+	if (!diff)
+		return;
+
 	if (diff > 0)
 		diff -= fill_balloon(vb, diff);
-	else if (diff < 0)
+	else
 		diff += leak_balloon(vb, -diff);
 	update_balloon_size(vb);
 
-- 
2.7.4

Re: [PATCH v1 0/2] virtio-balloon: tweak config_changed

[Cornelia Huck]

On Thu,  3 Jan 2019 13:31:00 +0800
Wei Wang <wei.w.wang@intel.com> wrote:

> Since virtio-ccw doesn't work with accessing to the config registers
> inside an interrupt context, this patch series avoids that issue by
> moving the config register accesses to the related workqueue contexts.
> 
> Wei Wang (2):
>   virtio-balloon: tweak config_changed implementation
>   virtio-balloon: improve update_balloon_size_func
> 
>  drivers/virtio/virtio_balloon.c | 59 +++++++++++++++++++++--------------------
>  1 file changed, 30 insertions(+), 29 deletions(-)
> 

A quick test (various 'balloon <num>' in the QEMU monitor) with a
virtio-balloon-ccw device seems to work fine here.

Re: [PATCH v1 1/2] virtio-balloon: tweak config_changed implementation

[Cornelia Huck]

On Thu,  3 Jan 2019 13:31:01 +0800
Wei Wang <wei.w.wang@intel.com> wrote:

> virtio-ccw has deadlock issues with reading config registers inside the

s/config registers/the config space/ ?

> interrupt context, so we tweak the virtballoon_changed implementation
> by moving the config read operations into the related workqueue contexts.

Also credit Christian with a Reported-by:?

> 
> Signed-off-by: Wei Wang <wei.w.wang@intel.com>
> ---
>  drivers/virtio/virtio_balloon.c | 54 ++++++++++++++++++++---------------------
>  1 file changed, 26 insertions(+), 28 deletions(-)

Reviewed-by: Cornelia Huck <cohuck@redhat.com>

Re: [PATCH v1 2/2] virtio-balloon: improve update_balloon_size_func

[Cornelia Huck]

On Thu,  3 Jan 2019 13:31:02 +0800
Wei Wang <wei.w.wang@intel.com> wrote:

> There is no need to update the balloon actual register when there is no
> ballooning request. This patch avoids update_balloon_size when diff is 0.
> 
> Signed-off-by: Wei Wang <wei.w.wang@intel.com>
> ---
>  drivers/virtio/virtio_balloon.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)

Reviewed-by: Cornelia Huck <cohuck@redhat.com>

Re: [PATCH v1 1/2] virtio-balloon: tweak config_changed implementation

[Wei Wang]

On 01/03/2019 05:40 PM, Cornelia Huck wrote:
> On Thu,  3 Jan 2019 13:31:01 +0800
> Wei Wang <wei.w.wang@intel.com> wrote:
>
>> virtio-ccw has deadlock issues with reading config registers inside the
> s/config registers/the config space/ ?

Sounds good.

>
>> interrupt context, so we tweak the virtballoon_changed implementation
>> by moving the config read operations into the related workqueue contexts.
> Also credit Christian with a Reported-by:?

Yes, definitely. Sorry for missing that.

Best,
Wei

Re: [PATCH v1 1/2] virtio-balloon: tweak config_changed implementation

[Halil Pasic]

On Thu,  3 Jan 2019 13:31:01 +0800
Wei Wang <wei.w.wang@intel.com> wrote:

> virtio-ccw has deadlock issues with reading config registers inside the
> interrupt context

I would say something like 'virtio-ccw does not support using
virtio_config_ops from an atomic context' as the limitation is not
limited to read config (aka. get()). But I'm fine with your formulation
as well.

> , so we tweak the virtballoon_changed implementation
> by moving the config read operations into the related workqueue contexts.
> 
> Signed-off-by: Wei Wang <wei.w.wang@intel.com>


Reviewed-by: Halil Pasic <pasic@linux.ibm.com>

[..]

Re: [PATCH v1 2/2] virtio-balloon: improve update_balloon_size_func

[Halil Pasic]

On Thu,  3 Jan 2019 13:31:02 +0800
Wei Wang <wei.w.wang@intel.com> wrote:

> There is no need to update the balloon actual register when there is no
> ballooning request. This patch avoids update_balloon_size when diff is 0.
> 
> Signed-off-by: Wei Wang <wei.w.wang@intel.com>

Reviewed-by: Halil Pasic <pasic@linux.ibm.com>

I would be also fine with both patches squashed into one.

BTW the end result looks to me even a bit nicer.

Re: [PATCH v1 1/2] virtio-balloon: tweak config_changed implementation

[Michael S. Tsirkin]

On Thu, Jan 03, 2019 at 01:31:01PM +0800, Wei Wang wrote:
> virtio-ccw has deadlock issues with reading config registers inside the
> interrupt context, so we tweak the virtballoon_changed implementation
> by moving the config read operations into the related workqueue contexts.
> 
> Signed-off-by: Wei Wang <wei.w.wang@intel.com>
> ---
>  drivers/virtio/virtio_balloon.c | 54 ++++++++++++++++++++---------------------
>  1 file changed, 26 insertions(+), 28 deletions(-)
> 
> diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
> index 728ecd1..9a82a11 100644
> --- a/drivers/virtio/virtio_balloon.c
> +++ b/drivers/virtio/virtio_balloon.c
> @@ -394,33 +394,15 @@ static void virtballoon_changed(struct virtio_device *vdev)
>  {
>  	struct virtio_balloon *vb = vdev->priv;
>  	unsigned long flags;
> -	s64 diff = towards_target(vb);
> -
> -	if (diff) {
> -		spin_lock_irqsave(&vb->stop_update_lock, flags);
> -		if (!vb->stop_update)
> -			queue_work(system_freezable_wq,
> -				   &vb->update_balloon_size_work);
> -		spin_unlock_irqrestore(&vb->stop_update_lock, flags);
> -	}
>  
> -	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
> -		virtio_cread(vdev, struct virtio_balloon_config,
> -			     free_page_report_cmd_id, &vb->cmd_id_received);

There's one problem with this approach:

previously updating the cmd_id_received here would immediately
stop the report in send_free_pages.

With this approach we are waiting for the wq to schedule,
which might be blocked waiting for report to complete.
So host can no longer quickly stop the report in progress.

A simple work-around would be to set some kind of flag whenever there
is a change interrupt, then have send_free_pages test it
and re-read cmd_id_received.

Needs to be an atomic I guess ...


> -		if (vb->cmd_id_received == VIRTIO_BALLOON_CMD_ID_DONE) {
> -			/* Pass ULONG_MAX to give back all the free pages */
> -			return_free_pages_to_mm(vb, ULONG_MAX);
> -		} else if (vb->cmd_id_received != VIRTIO_BALLOON_CMD_ID_STOP &&
> -			   vb->cmd_id_received !=
> -			   virtio32_to_cpu(vdev, vb->cmd_id_active)) {
> -			spin_lock_irqsave(&vb->stop_update_lock, flags);
> -			if (!vb->stop_update) {
> -				queue_work(vb->balloon_wq,
> -					   &vb->report_free_page_work);
> -			}
> -			spin_unlock_irqrestore(&vb->stop_update_lock, flags);
> -		}
> +	spin_lock_irqsave(&vb->stop_update_lock, flags);
> +	if (!vb->stop_update) {
> +		queue_work(system_freezable_wq,
> +			   &vb->update_balloon_size_work);
> +		if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
> +			queue_work(vb->balloon_wq, &vb->report_free_page_work);
>  	}
> +	spin_unlock_irqrestore(&vb->stop_update_lock, flags);
>  }
>  
>  static void update_balloon_size(struct virtio_balloon *vb)
> @@ -637,11 +619,9 @@ static int send_free_pages(struct virtio_balloon *vb)
>  	return 0;
>  }
>  
> -static void report_free_page_func(struct work_struct *work)
> +static void virtio_balloon_report_free_page(struct virtio_balloon *vb)
>  {
>  	int err;
> -	struct virtio_balloon *vb = container_of(work, struct virtio_balloon,
> -						 report_free_page_work);
>  	struct device *dev = &vb->vdev->dev;
>  
>  	/* Start by sending the received cmd id to host with an outbuf. */
> @@ -659,6 +639,24 @@ static void report_free_page_func(struct work_struct *work)
>  		dev_err(dev, "Failed to send a stop id, err = %d\n", err);
>  }
>  
> +static void report_free_page_func(struct work_struct *work)
> +{
> +	struct virtio_balloon *vb = container_of(work, struct virtio_balloon,
> +						 report_free_page_work);
> +
> +	virtio_cread(vb->vdev, struct virtio_balloon_config,
> +		     free_page_report_cmd_id, &vb->cmd_id_received);
> +
> +	if (vb->cmd_id_received == VIRTIO_BALLOON_CMD_ID_DONE) {
> +		/* Pass ULONG_MAX to give back all the free pages */
> +		return_free_pages_to_mm(vb, ULONG_MAX);
> +	} else if (vb->cmd_id_received != VIRTIO_BALLOON_CMD_ID_STOP &&
> +		   vb->cmd_id_received !=
> +		   virtio32_to_cpu(vb->vdev, vb->cmd_id_active)) {
> +		virtio_balloon_report_free_page(vb);
> +	}
> +}
> +
>  #ifdef CONFIG_BALLOON_COMPACTION
>  /*
>   * virtballoon_migratepage - perform the balloon page migration on behalf of
> -- 
> 2.7.4

Re: [virtio-dev] Re: [PATCH v1 1/2] virtio-balloon: tweak config_changed implementation

[Wei Wang]

On 01/04/2019 12:42 AM, Michael S. Tsirkin wrote:
> On Thu, Jan 03, 2019 at 01:31:01PM +0800, Wei Wang wrote:
>> virtio-ccw has deadlock issues with reading config registers inside the
>> interrupt context, so we tweak the virtballoon_changed implementation
>> by moving the config read operations into the related workqueue contexts.
>>
>> Signed-off-by: Wei Wang <wei.w.wang@intel.com>
>> ---
>>   drivers/virtio/virtio_balloon.c | 54 ++++++++++++++++++++---------------------
>>   1 file changed, 26 insertions(+), 28 deletions(-)
>>
>> diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
>> index 728ecd1..9a82a11 100644
>> --- a/drivers/virtio/virtio_balloon.c
>> +++ b/drivers/virtio/virtio_balloon.c
>> @@ -394,33 +394,15 @@ static void virtballoon_changed(struct virtio_device *vdev)
>>   {
>>   	struct virtio_balloon *vb = vdev->priv;
>>   	unsigned long flags;
>> -	s64 diff = towards_target(vb);
>> -
>> -	if (diff) {
>> -		spin_lock_irqsave(&vb->stop_update_lock, flags);
>> -		if (!vb->stop_update)
>> -			queue_work(system_freezable_wq,
>> -				   &vb->update_balloon_size_work);
>> -		spin_unlock_irqrestore(&vb->stop_update_lock, flags);
>> -	}
>>   
>> -	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
>> -		virtio_cread(vdev, struct virtio_balloon_config,
>> -			     free_page_report_cmd_id, &vb->cmd_id_received);
> There's one problem with this approach:
>
> previously updating the cmd_id_received here would immediately
> stop the report in send_free_pages.
>
> With this approach we are waiting for the wq to schedule,
> which might be blocked waiting for report to complete.
> So host can no longer quickly stop the report in progress.
>
> A simple work-around would be to set some kind of flag whenever there
> is a change interrupt, then have send_free_pages test it
> and re-read cmd_id_received.
>
> Needs to be an atomic I guess ...
>

Yes, sounds better..will update the patch.

Best,
Wei