diff mbox series

[v2,1/4] vhost: enforce avail index and desc read ordering

Message ID 20181219082113.24455-2-maxime.coquelin@redhat.com (mailing list archive)
State Accepted, archived
Delegated to: Maxime Coquelin
Headers show
Series vhost: add missing barriers, move prefetching | expand

Checks

Context Check Description
ci/Intel-compilation success Compilation OK
ci/mellanox-Performance-Testing success Performance Testing PASS
ci/intel-Performance-Testing success Performance Testing PASS

Commit Message

Maxime Coquelin Dec. 19, 2018, 8:21 a.m. UTC
A read barrier is required to ensure the ordering between
available index and the descriptor reads is enforced.

1. read avail_head = avail->idx
2. read cur_idx = last_avail_idx
if (cur_idx != avail_head) {
    3. read idx = avail->ring[cur_idx]
    4. read desc[idx]
}

There is a control dependency between step 1 and steps 3 & 4,
3 could be speculatively executed before 1, which could result
in 'idx' to not being updated yet.

Fixes: 4796ad63ba1f ("examples/vhost: import userspace vhost application")
Cc: stable@dpdk.org

Reported-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Acked-by: Ilya Maximets <i.maximets@samsung.com>
---
 lib/librte_vhost/virtio_net.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

Comments

Michael S. Tsirkin Dec. 19, 2018, 3:47 p.m. UTC | #1
On Wed, Dec 19, 2018 at 09:21:10AM +0100, Maxime Coquelin wrote:
> A read barrier is required to ensure the ordering between
> available index and the descriptor reads is enforced.
> 
> 1. read avail_head = avail->idx
> 2. read cur_idx = last_avail_idx
> if (cur_idx != avail_head) {
>     3. read idx = avail->ring[cur_idx]
>     4. read desc[idx]
> }
> 
> There is a control dependency between step 1 and steps 3 & 4,
> 3 could be speculatively executed before 1, which could result
> in 'idx' to not being updated yet.
> 
> Fixes: 4796ad63ba1f ("examples/vhost: import userspace vhost application")
> Cc: stable@dpdk.org
> 
> Reported-by: Jason Wang <jasowang@redhat.com>
> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> Acked-by: Ilya Maximets <i.maximets@samsung.com>

Acked-by: Michael S. Tsirkin <mst@redhat.com>

BTW Ilya do you see a performance degradation from RMBs with these patches?


> ---
>  lib/librte_vhost/virtio_net.c | 12 ++++++++++++
>  1 file changed, 12 insertions(+)
> 
> diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
> index 8c657a101..7f37bbbed 100644
> --- a/lib/librte_vhost/virtio_net.c
> +++ b/lib/librte_vhost/virtio_net.c
> @@ -752,6 +752,12 @@ virtio_dev_rx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
>  	rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
>  	avail_head = *((volatile uint16_t *)&vq->avail->idx);
>  
> +	/*
> +	 * The ordering between avail index and
> +	 * desc reads needs to be enforced.
> +	 */
> +	rte_smp_rmb();
> +

I'd guess you want to put the RMB before the prefetch. No?
Otherwise I think you are either stalling until that completes
or discard the prefetch ...


>  	for (pkt_idx = 0; pkt_idx < count; pkt_idx++) {
>  		uint32_t pkt_len = pkts[pkt_idx]->pkt_len + dev->vhost_hlen;
>  		uint16_t nr_vec = 0;
> @@ -1334,6 +1340,12 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
>  	if (free_entries == 0)
>  		return 0;
>  
> +	/*
> +	 * The ordering between avail index and
> +	 * desc reads needs to be enforced.
> +	 */
> +	rte_smp_rmb();
> +
>  	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
>  
>  	count = RTE_MIN(count, MAX_PKT_BURST);
> -- 
> 2.17.2
Maxime Coquelin Dec. 19, 2018, 3:50 p.m. UTC | #2
On 12/19/18 4:47 PM, Michael S. Tsirkin wrote:
> On Wed, Dec 19, 2018 at 09:21:10AM +0100, Maxime Coquelin wrote:
>> A read barrier is required to ensure the ordering between
>> available index and the descriptor reads is enforced.
>>
>> 1. read avail_head = avail->idx
>> 2. read cur_idx = last_avail_idx
>> if (cur_idx != avail_head) {
>>      3. read idx = avail->ring[cur_idx]
>>      4. read desc[idx]
>> }
>>
>> There is a control dependency between step 1 and steps 3 & 4,
>> 3 could be speculatively executed before 1, which could result
>> in 'idx' to not being updated yet.
>>
>> Fixes: 4796ad63ba1f ("examples/vhost: import userspace vhost application")
>> Cc: stable@dpdk.org
>>
>> Reported-by: Jason Wang <jasowang@redhat.com>
>> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
>> Acked-by: Ilya Maximets <i.maximets@samsung.com>
> 
> Acked-by: Michael S. Tsirkin <mst@redhat.com>
> 
> BTW Ilya do you see a performance degradation from RMBs with these patches?
> 
> 
>> ---
>>   lib/librte_vhost/virtio_net.c | 12 ++++++++++++
>>   1 file changed, 12 insertions(+)
>>
>> diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
>> index 8c657a101..7f37bbbed 100644
>> --- a/lib/librte_vhost/virtio_net.c
>> +++ b/lib/librte_vhost/virtio_net.c
>> @@ -752,6 +752,12 @@ virtio_dev_rx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
>>   	rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
>>   	avail_head = *((volatile uint16_t *)&vq->avail->idx);
>>   
>> +	/*
>> +	 * The ordering between avail index and
>> +	 * desc reads needs to be enforced.
>> +	 */
>> +	rte_smp_rmb();
>> +
> 
> I'd guess you want to put the RMB before the prefetch. No?
> Otherwise I think you are either stalling until that completes
> or discard the prefetch ...

Yeah, this is actually done in patch 3.

Thanks,
Maxime
>>   	for (pkt_idx = 0; pkt_idx < count; pkt_idx++) {
>>   		uint32_t pkt_len = pkts[pkt_idx]->pkt_len + dev->vhost_hlen;
>>   		uint16_t nr_vec = 0;
>> @@ -1334,6 +1340,12 @@ virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
>>   	if (free_entries == 0)
>>   		return 0;
>>   
>> +	/*
>> +	 * The ordering between avail index and
>> +	 * desc reads needs to be enforced.
>> +	 */
>> +	rte_smp_rmb();
>> +
>>   	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
>>   
>>   	count = RTE_MIN(count, MAX_PKT_BURST);
>> -- 
>> 2.17.2
diff mbox series

Patch

diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
index 8c657a101..7f37bbbed 100644
--- a/lib/librte_vhost/virtio_net.c
+++ b/lib/librte_vhost/virtio_net.c
@@ -752,6 +752,12 @@  virtio_dev_rx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
 	avail_head = *((volatile uint16_t *)&vq->avail->idx);
 
+	/*
+	 * The ordering between avail index and
+	 * desc reads needs to be enforced.
+	 */
+	rte_smp_rmb();
+
 	for (pkt_idx = 0; pkt_idx < count; pkt_idx++) {
 		uint32_t pkt_len = pkts[pkt_idx]->pkt_len + dev->vhost_hlen;
 		uint16_t nr_vec = 0;
@@ -1334,6 +1340,12 @@  virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	if (free_entries == 0)
 		return 0;
 
+	/*
+	 * The ordering between avail index and
+	 * desc reads needs to be enforced.
+	 */
+	rte_smp_rmb();
+
 	VHOST_LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
 
 	count = RTE_MIN(count, MAX_PKT_BURST);