Hi Yuanhan,
On 6/6/2016 4:01 PM, Yuanhan Liu wrote: > On Thu, Jun 02, 2016 at 09:54:36AM +0000, Jianfeng Tan wrote: >> + >> + desc_addr = (uint64_t)vq->mz->addr; >> + avail_addr = desc_addr + vq->vq_nentries * sizeof(struct vring_desc); >> + used_addr = RTE_ALIGN_CEIL(avail_addr + offsetof(struct vring_avail, >> + ring[vq->vq_nentries]), >> + VIRTIO_PCI_VRING_ALIGN); >> + >> + dev->vrings[queue_idx].num = vq->vq_nentries; >> + dev->vrings[queue_idx].desc = (void *)desc_addr; >> + dev->vrings[queue_idx].avail = (void *)avail_addr; >> + dev->vrings[queue_idx].used = (void *)used_addr; > That would break 32 bit build. please also do more build and function > test, with and without CONFIG_RTE_VIRTIO_VDEV enabled, to make sure > we will not break anything. I'm sure you will meet build error without > that option enabled. Yes, thanks for pointing this out. > > BTW, let's be consistent with using VIRTIO_USER_DEV instead of VDEV > or VIRTIO_VDEV? OK. > > Another thing that might be a bit late to ask is that how about > removing the vhost-net support? I mean, it's DPDK; if user stick > to using DPDK virtio-user, he will stick to using DPDK vhost-user > as well, but not the vhost-net. So, let's keep it being simple > first. And if there is really a need for vhost-net, we can add it > back later, easily. Makes sense? Yes, it makes sense, because from an initial test, I see low performance. Or anyone who are willing to use it can comment? Thanks, Jianfeng > > I also would suggest you do a rebase based on my latest tree. No problem. Thanks, Jianfeng > > --yliu