Re: [PATCH v3 2/6] vhost-user: Enable multiqueue

[Laurent Vivier <lvivier@redhat.com>]

On 12/11/25 08:01, Stefano Brivio wrote:
> On Wed,  3 Dec 2025 19:54:30 +0100
> Laurent Vivier <lvivier@redhat.com> wrote:
> 
>> Advertise multi-queue support in vhost-user by setting VIRTIO_NET_F_MQ
>> and VHOST_USER_PROTOCOL_F_MQ feature flags, and increase
>> VHOST_USER_MAX_VQS from 2 to 32, supporting up to 16 queue pairs.
>>
>> Currently, only the first RX queue (queue 0) is used for receiving
>> packets. The guest kernel selects which TX queue to use for
>> transmission. Full multi-RX queue load balancing will be implemented in
>> future work.
>>
>> Update the QEMU usage hint to show the required parameters for enabling
>> multiqueue: queues parameter on the netdev, and mq=true on the
>> virtio-net device.
>>
>> Signed-off-by: Laurent Vivier <lvivier@redhat.com>
>> ---
>>   tap.c        |  7 +++++--
>>   vhost_user.c | 10 ++++++----
>>   virtio.h     |  2 +-
>>   3 files changed, 12 insertions(+), 7 deletions(-)
>>
>> diff --git a/tap.c b/tap.c
>> index 2cda8c9772b8..591b49491aa3 100644
>> --- a/tap.c
>> +++ b/tap.c
>> @@ -1314,8 +1314,11 @@ static void tap_backend_show_hints(struct ctx *c)
>>   		break;
>>   	case MODE_VU:
>>   		info("You can start qemu with:");
>> -		info("    kvm ... -chardev socket,id=chr0,path=%s -netdev vhost-user,id=netdev0,chardev=chr0 -device virtio-net,netdev=netdev0 -object memory-backend-memfd,id=memfd0,share=on,size=$RAMSIZE -numa node,memdev=memfd0\n",
>> -		     c->sock_path);
>> +		info("    kvm ... -chardev socket,id=chr0,path=%s "
>> +		     "-netdev vhost-user,id=netdev0,chardev=chr0,queues=$QUEUES "
>> +		     "-device virtio-net,netdev=netdev0,mq=true "
>> +		     "-object memory-backend-memfd,id=memfd0,share=on,size=$RAMSIZE "
>> +		     "-numa node,memdev=memfd0\n", c->sock_path);
>>   		break;
>>   	}
>>   }
>> diff --git a/vhost_user.c b/vhost_user.c
>> index aa7c869d9e56..845fdb551c84 100644
>> --- a/vhost_user.c
>> +++ b/vhost_user.c
>> @@ -323,6 +323,7 @@ static bool vu_get_features_exec(struct vu_dev *vdev,
>>   	uint64_t features =
>>   		1ULL << VIRTIO_F_VERSION_1 |
>>   		1ULL << VIRTIO_NET_F_MRG_RXBUF |
>> +		1ULL << VIRTIO_NET_F_MQ |
>>   		1ULL << VHOST_F_LOG_ALL |
>>   		1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
>>   
>> @@ -767,7 +768,8 @@ static void vu_check_queue_msg_file(struct vhost_user_msg *vmsg)
>>   	int idx = vmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
>>   
>>   	if (idx >= VHOST_USER_MAX_VQS)
>> -		die("Invalid vhost-user queue index: %u", idx);
>> +		die("Invalid vhost-user queue index: %u (maximum %u)", idx,
>> +		    VHOST_USER_MAX_VQS);
>>   
>>   	if (nofd) {
>>   		vmsg_close_fds(vmsg);
>> @@ -896,7 +898,8 @@ static bool vu_get_protocol_features_exec(struct vu_dev *vdev,
>>   	uint64_t features = 1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK |
>>   			    1ULL << VHOST_USER_PROTOCOL_F_LOG_SHMFD |
>>   			    1ULL << VHOST_USER_PROTOCOL_F_DEVICE_STATE |
>> -			    1ULL << VHOST_USER_PROTOCOL_F_RARP;
>> +			    1ULL << VHOST_USER_PROTOCOL_F_RARP |
>> +			    1ULL << VHOST_USER_PROTOCOL_F_MQ;
>>   
>>   	(void)vdev;
>>   	vmsg_set_reply_u64(vmsg, features);
>> @@ -935,10 +938,9 @@ static bool vu_get_queue_num_exec(struct vu_dev *vdev,
>>   {
>>   	(void)vdev;
>>   
>> -	/* NOLINTNEXTLINE(misc-redundant-expression) */
>>   	vmsg_set_reply_u64(vmsg, VHOST_USER_MAX_VQS / 2);
>>   
>> -	debug("VHOST_USER_MAX_VQS  %u", VHOST_USER_MAX_VQS / 2);
>> +	debug("queue num  %u", VHOST_USER_MAX_VQS / 2);
> 
> Nit, if you respin: this "queue num  %u" message doesn't carry any
> context at all. Actually, why is it needed? It's defined at build time
> anyway.
> 
> If it's needed maybe "Using up to %u vhost-user queue pairs"?

I agree, and I was planning to update it but missed the change before posting.

Thanks,
Laurent