From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:34785) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1fWG8v-0008VT-AZ for qemu-devel@nongnu.org; Fri, 22 Jun 2018 03:06:02 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1fWG8r-00057B-Cu for qemu-devel@nongnu.org; Fri, 22 Jun 2018 03:06:01 -0400 References: <90a232ff-74f7-8d10-74f4-87a0008c41df@redhat.com> From: Jia He Message-ID: Date: Fri, 22 Jun 2018 15:05:51 +0800 MIME-Version: 1.0 In-Reply-To: <90a232ff-74f7-8d10-74f4-87a0008c41df@redhat.com> Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [Questio] What the proper vfio NIC for emulation smmu? List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Auger Eric Cc: qemu-arm@nongnu.org, qemu-devel@nongnu.org Hi Eric On 6/22/2018 2:39 PM, Auger Eric Wrote: > Hello Jia, > > On 06/22/2018 03:44 AM, Jia He wrote: >> Hi Eric >> I want to test your smmu emulation patchset with vfio-pci ethernet device. >> Could you kindly give me the NIC type information. >> In my test, there is vf probe error for menallox connect 4 NIC and TX Unit >> hang error for Intel X540 NIC. That's the reason why I ask this question. >> > virtual SMMUv3 is not integrated with VFIO and is not planned to be. It > was tested with virtio-net-pci (without vhost-net). > > With upstream code you should see the following warning: > > "SMMUV3 does not support vhost/vfio integration yet: devices of those > types will not function properly" > > With "[PATCH v3 0/4] ARM SMMUv3: IOTLB Emulation and VHOST Support", > vhost-net gets integrated. VFIO won't be integrated either. If vfio-pci > device are attempted to be used along with vsmmu, you will get > "SMMUv3 does not support notification on MAP: device %s will not > function properly" Thanks for the information. Actually, I am using your vsmmu-optim-v3 branch. I assumed vfio worked fine since you removed that warning in commit ("hw/arm/smmuv3: Add notifications on invalidation"). Seems I am wrong ;-) > > My attempt to integrate the vsmmu with VFIO in the same manner as intel > iommu (using a fake caching mode) was rejected by the kernel maintainer. > So for VFIO use case, we need either 2 stage enablement (not implemented > yet) or virtio-iommu. I will respin this latter against v0.7 spec > probably next week. Ok. got it, thanks > > Thanks > > Eric > -- Cheers, Jia