[PATCH v8 06/13] vdpa/ifc: add block device SW live-migration

Xia, Chenbo chenbo.xia at intel.com
Mon May 23 07:25:26 CEST 2022


> -----Original Message-----
> From: Pei, Andy <andy.pei at intel.com>
> Sent: Wednesday, May 18, 2022 8:14 PM
> To: dev at dpdk.org
> Cc: Xia, Chenbo <chenbo.xia at intel.com>; maxime.coquelin at redhat.com; Cao,
> Gang <gang.cao at intel.com>; Liu, Changpeng <changpeng.liu at intel.com>; Xu,
> Rosen <rosen.xu at intel.com>; Xiao, QimaiX <qimaix.xiao at intel.com>
> Subject: [PATCH v8 06/13] vdpa/ifc: add block device SW live-migration
> 
> Add SW live-migration support to block device.
> For block device, it is critical that no packet
> should be dropped. So when virtio blk device is
> pause, make sure hardware last_avail_idx and

paused

> last_used_idx are the same. This indicates all
> requests have recieved acks, and no inflight IO.

received

With above fixed:

Reviewed-by: Chenbo Xia <chenbo.xia at intel.com>

> 
> Signed-off-by: Andy Pei <andy.pei at intel.com>
> ---
>  drivers/vdpa/ifc/base/ifcvf.h |  1 +
>  drivers/vdpa/ifc/ifcvf_vdpa.c | 42
> ++++++++++++++++++++++++++++++++++++++----
>  2 files changed, 39 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/vdpa/ifc/base/ifcvf.h b/drivers/vdpa/ifc/base/ifcvf.h
> index 244de46..4fb1736 100644
> --- a/drivers/vdpa/ifc/base/ifcvf.h
> +++ b/drivers/vdpa/ifc/base/ifcvf.h
> @@ -65,6 +65,7 @@
>  #define IFCVF_MEDIATED_VRING		0x200000000000
> 
>  #define IFCVF_32_BIT_MASK		0xffffffff
> +#define IFCVF_16_BIT_MASK		0xffff
> 
> 
>  #ifndef VHOST_USER_PROTOCOL_F_CONFIG
> diff --git a/drivers/vdpa/ifc/ifcvf_vdpa.c b/drivers/vdpa/ifc/ifcvf_vdpa.c
> index 509a1ed..3e78c7d 100644
> --- a/drivers/vdpa/ifc/ifcvf_vdpa.c
> +++ b/drivers/vdpa/ifc/ifcvf_vdpa.c
> @@ -316,8 +316,34 @@ struct rte_vdpa_dev_info {
>  	uint64_t features = 0;
>  	uint64_t log_base = 0, log_size = 0;
>  	uint64_t len;
> +	u32 ring_state = 0;
> 
>  	vid = internal->vid;
> +
> +	/* to make sure no packet is lost for blk device
> +	 * do not stop until last_avail_idx == last_used_idx
> +	 */
> +	if (internal->hw.device_type == IFCVF_BLK) {
> +		for (i = 0; i < hw->nr_vring; i++) {
> +			do {
> +				if (hw->lm_cfg != NULL)
> +					ring_state = *(u32 *)(hw->lm_cfg +
> +						IFCVF_LM_RING_STATE_OFFSET +
> +						i * IFCVF_LM_CFG_SIZE);
> +				hw->vring[i].last_avail_idx =
> +					(u16)(ring_state & IFCVF_16_BIT_MASK);
> +				hw->vring[i].last_used_idx =
> +					(u16)(ring_state >> 16);
> +				if (hw->vring[i].last_avail_idx !=
> +					hw->vring[i].last_used_idx) {
> +					ifcvf_notify_queue(hw, i);
> +					usleep(10);
> +				}
> +			} while (hw->vring[i].last_avail_idx !=
> +				hw->vring[i].last_used_idx);
> +		}
> +	}
> +
>  	ifcvf_stop_hw(hw);
> 
>  	for (i = 0; i < hw->nr_vring; i++)
> @@ -641,8 +667,10 @@ struct rte_vdpa_dev_info {
>  		}
>  		hw->vring[i].avail = gpa;
> 
> -		/* Direct I/O for Tx queue, relay for Rx queue */
> -		if (i & 1) {
> +		/* NET: Direct I/O for Tx queue, relay for Rx queue
> +		 * BLK: relay every queue
> +		 */
> +		if ((internal->hw.device_type == IFCVF_NET) && (i & 1)) {
>  			gpa = hva_to_gpa(vid, (uint64_t)(uintptr_t)vq.used);
>  			if (gpa == 0) {
>  				DRV_LOG(ERR, "Fail to get GPA for used ring.");
> @@ -692,8 +720,12 @@ struct rte_vdpa_dev_info {
> 
>  	for (i = 0; i < hw->nr_vring; i++) {
>  		/* synchronize remaining new used entries if any */
> -		if ((i & 1) == 0)
> +		if (internal->hw.device_type == IFCVF_NET) {
> +			if ((i & 1) == 0)
> +				update_used_ring(internal, i);
> +		} else if (internal->hw.device_type == IFCVF_BLK) {
>  			update_used_ring(internal, i);
> +		}
> 
>  		rte_vhost_get_vhost_vring(vid, i, &vq);
>  		len = IFCVF_USED_RING_LEN(vq.size);
> @@ -755,7 +787,9 @@ struct rte_vdpa_dev_info {
>  		}
>  	}
> 
> -	for (qid = 0; qid < q_num; qid += 2) {
> +	for (qid = 0; qid < q_num; qid += 1) {
> +		if ((internal->hw.device_type == IFCVF_NET) && (qid & 1))
> +			continue;
>  		ev.events = EPOLLIN | EPOLLPRI;
>  		/* leave a flag to mark it's for interrupt */
>  		ev.data.u64 = 1 | qid << 1 |
> --
> 1.8.3.1



More information about the dev mailing list