[PATCH v3 2/2] examples/vhost: unconfigure DMA vchannel

[Ding, Xuan]

Hi Chenbo,

> -----Original Message-----
> From: Xia, Chenbo <chenbo.xia at intel.com>
> Sent: Thursday, September 29, 2022 4:28 PM
> To: Ding, Xuan <xuan.ding at intel.com>; maxime.coquelin at redhat.com
> Cc: dev at dpdk.org; Hu, Jiayu <jiayu.hu at intel.com>; He, Xingguang
> <xingguang.he at intel.com>; Yang, YvonneX <yvonnex.yang at intel.com>;
> Jiang, Cheng1 <cheng1.jiang at intel.com>; Wang, YuanX
> <yuanx.wang at intel.com>; Ma, WenwuX <wenwux.ma at intel.com>
> Subject: RE: [PATCH v3 2/2] examples/vhost: unconfigure DMA vchannel
> 
> > -----Original Message-----
> > From: Ding, Xuan <xuan.ding at intel.com>
> > Sent: Thursday, September 29, 2022 9:33 AM
> > To: maxime.coquelin at redhat.com; Xia, Chenbo <chenbo.xia at intel.com>
> > Cc: dev at dpdk.org; Hu, Jiayu <jiayu.hu at intel.com>; He, Xingguang
> > <xingguang.he at intel.com>; Yang, YvonneX <yvonnex.yang at intel.com>;
> > Jiang,
> > Cheng1 <cheng1.jiang at intel.com>; Wang, YuanX
> <yuanx.wang at intel.com>;
> > Ma, WenwuX <wenwux.ma at intel.com>; Ding, Xuan
> <xuan.ding at intel.com>
> > Subject: [PATCH v3 2/2] examples/vhost: unconfigure DMA vchannel
> >
> > From: Xuan Ding <xuan.ding at intel.com>
> >
> > This patch applies rte_vhost_async_dma_unconfigure() API to manually
> > free DMA vchannels instead of waiting until the program ends to be
> > released.
> >
> > Signed-off-by: Xuan Ding <xuan.ding at intel.com>
> > ---
> >  examples/vhost/main.c | 45
> > ++++++++++++++++++++++++++++++-------------
> >  examples/vhost/main.h |  1 +
> >  2 files changed, 33 insertions(+), 13 deletions(-)
> >
> > diff --git a/examples/vhost/main.c b/examples/vhost/main.c index
> > 0fa4753e70..32f396d88a 100644
> > --- a/examples/vhost/main.c
> > +++ b/examples/vhost/main.c
> > @@ -1558,6 +1558,28 @@ vhost_clear_queue(struct vhost_dev *vdev,
> > uint16_t
> > queue_id)
> >  	}
> >  }
> >
> > +static void
> > +vhost_clear_async(struct vhost_dev *vdev, int vid, uint16_t queue_id)
> > +{
> > +	int16_t dma_id;
> > +	uint16_t ref_count;
> > +
> > +	if (dma_bind[vid].dmas[queue_id].async_enabled) {
> > +		vhost_clear_queue(vdev, queue_id);
> > +		rte_vhost_async_channel_unregister(vid, queue_id);
> > +		dma_bind[vid].dmas[queue_id].async_enabled = false;
> > +	}
> > +
> > +	dma_id = dma_bind[vid2socketid[vdev-
> >vid]].dmas[queue_id].dev_id;
> > +	dma_bind[vid2socketid[vdev-
> >vid]].dmas[queue_id].dma_ref_count--;
> > +	ref_count = dma_bind[vid2socketid[vdev-
> > >vid]].dmas[queue_id].dma_ref_count;
> 
> Above two lines could be combined like 'ref_count = --XXXX'
> 
> But I wonder is this dma_ref_count really needed? It's always 0/1 like
> Async_enabled, no? If it target at multi-port sharing same one dma case, it
> should not be implemented like this.

Yes, the dma_ref_count here is for multi-port sharing same one dma case. The intention was to
unconfigure dma device when it is no longer used by any ports.
I will check the implementation.

Thanks,
Xuan

> 
> Thanks,
> Chenbo
> 
> > +
> > +	if (ref_count == 0 && dma_id != INVALID_DMA_ID) {
> > +		if (rte_vhost_async_dma_unconfigure(dma_id, 0) < 0)
> > +			RTE_LOG(ERR, VHOST_PORT, "Failed to unconfigure
> DMA in
> > vhost.\n");
> > +	}
> > +}
> > +
> >  /*
> >   * Remove a device from the specific data core linked list and from the
> >   * main linked list. Synchronization  occurs through the use of the
> > @@ -1614,17 +1636,8 @@ destroy_device(int vid)
> >  		"(%d) device has been removed from data core\n",
> >  		vdev->vid);
> >
> > -	if (dma_bind[vid].dmas[VIRTIO_RXQ].async_enabled) {
> > -		vhost_clear_queue(vdev, VIRTIO_RXQ);
> > -		rte_vhost_async_channel_unregister(vid, VIRTIO_RXQ);
> > -		dma_bind[vid].dmas[VIRTIO_RXQ].async_enabled = false;
> > -	}
> > -
> > -	if (dma_bind[vid].dmas[VIRTIO_TXQ].async_enabled) {
> > -		vhost_clear_queue(vdev, VIRTIO_TXQ);
> > -		rte_vhost_async_channel_unregister(vid, VIRTIO_TXQ);
> > -		dma_bind[vid].dmas[VIRTIO_TXQ].async_enabled = false;
> > -	}
> > +	vhost_clear_async(vdev, vid, VIRTIO_RXQ);
> > +	vhost_clear_async(vdev, vid, VIRTIO_TXQ);
> >
> >  	rte_free(vdev);
> >  }
> > @@ -1673,14 +1686,19 @@ vhost_async_channel_register(int vid)
> >
> >  	if (dma_bind[vid2socketid[vid]].dmas[VIRTIO_RXQ].dev_id !=
> > INVALID_DMA_ID) {
> >  		rx_ret = rte_vhost_async_channel_register(vid,
> VIRTIO_RXQ);
> > -		if (rx_ret == 0)
> > +		if (rx_ret == 0) {
> >
> > 	dma_bind[vid2socketid[vid]].dmas[VIRTIO_RXQ].async_enabled =
> true;
> > +
> > 	dma_bind[vid2socketid[vid]].dmas[VIRTIO_RXQ].dma_ref_count++;
> > +		}
> > +
> >  	}
> >
> >  	if (dma_bind[vid2socketid[vid]].dmas[VIRTIO_TXQ].dev_id !=
> > INVALID_DMA_ID) {
> >  		tx_ret = rte_vhost_async_channel_register(vid,
> VIRTIO_TXQ);
> > -		if (tx_ret == 0)
> > +		if (tx_ret == 0) {
> >
> > 	dma_bind[vid2socketid[vid]].dmas[VIRTIO_TXQ].async_enabled =
> true;
> > +
> > 	dma_bind[vid2socketid[vid]].dmas[VIRTIO_TXQ].dma_ref_count++;
> > +		}
> >  	}
> >
> >  	return rx_ret | tx_ret;
> > @@ -1886,6 +1904,7 @@ reset_dma(void)
> >  		for (j = 0; j < RTE_MAX_QUEUES_PER_PORT * 2; j++) {
> >  			dma_bind[i].dmas[j].dev_id = INVALID_DMA_ID;
> >  			dma_bind[i].dmas[j].async_enabled = false;
> > +			dma_bind[i].dmas[j].dma_ref_count = 0;
> >  		}
> >  	}
> >
> > diff --git a/examples/vhost/main.h b/examples/vhost/main.h index
> > 2fcb8376c5..2b2cf828d3 100644
> > --- a/examples/vhost/main.h
> > +++ b/examples/vhost/main.h
> > @@ -96,6 +96,7 @@ struct dma_info {
> >  	struct rte_pci_addr addr;
> >  	int16_t dev_id;
> >  	bool async_enabled;
> > +	uint16_t dma_ref_count;
> >  };
> >
> >  struct dma_for_vhost {
> > --
> > 2.17.1