From: Peter Xu <peterx@redhat.com> To: Auger Eric <eric.auger@redhat.com> Cc: peter.maydell@linaro.org, kevin.tian@intel.com, mst@redhat.com, tn@semihalf.com, qemu-devel@nongnu.org, alex.williamson@redhat.com, qemu-arm@nongnu.org, jean-philippe@linaro.org, bharat.bhushan@nxp.com, eric.auger.pro@gmail.com Subject: Re: [Qemu-devel] [PATCH for-4.2 v10 03/15] virtio-iommu: Add skeleton Date: Fri, 30 Aug 2019 09:26:14 +0800 Message-ID: <20190830012614.GK8729@xz-x1> (raw) In-Reply-To: <f6a3b750-8439-7934-565b-37e1001f84d4@redhat.com> On Thu, Aug 29, 2019 at 02:18:42PM +0200, Auger Eric wrote: > Hi Peter, > > First of all, please forgive me for the delay. > On 8/15/19 3:54 PM, Peter Xu wrote: > > On Tue, Jul 30, 2019 at 07:21:25PM +0200, Eric Auger wrote: > >> +static void virtio_iommu_handle_command(VirtIODevice *vdev, VirtQueue *vq) > >> +{ > >> + VirtIOIOMMU *s = VIRTIO_IOMMU(vdev); > >> + struct virtio_iommu_req_head head; > >> + struct virtio_iommu_req_tail tail; > > > > [1] > > > >> + VirtQueueElement *elem; > >> + unsigned int iov_cnt; > >> + struct iovec *iov; > >> + size_t sz; > >> + > >> + for (;;) { > >> + elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); > >> + if (!elem) { > >> + return; > >> + } > >> + > >> + if (iov_size(elem->in_sg, elem->in_num) < sizeof(tail) || > >> + iov_size(elem->out_sg, elem->out_num) < sizeof(head)) { > >> + virtio_error(vdev, "virtio-iommu bad head/tail size"); > >> + virtqueue_detach_element(vq, elem, 0); > >> + g_free(elem); > >> + break; > >> + } > >> + > >> + iov_cnt = elem->out_num; > >> + iov = g_memdup(elem->out_sg, sizeof(struct iovec) * elem->out_num); > > > > Could I ask why memdup is needed here? > Indeed I don't think it is needed and besides iov is not freed! > > I got inspired from hw/net/virtio-net.c. To be honest I don't get why > the g_memdup is needed there either. The out_sg gets duplicated and > commands work on the duplicated data and not in place. Oh true, I found that it's because of calling of iov_discard_front(). Please have a look at 771b6ed37e3. Though it seems to me that virtio-iommu does not truncate iovs so it should not be needed. > > > >> + sz = iov_to_buf(iov, iov_cnt, 0, &head, sizeof(head)); > >> + if (unlikely(sz != sizeof(head))) { > >> + tail.status = VIRTIO_IOMMU_S_DEVERR; > > > > Do you need to zero the reserved bits to make sure it won't contain > > garbage? Same question to below uses of tail. > yes. I initialized tail. > > > >> + goto out; > >> + } > >> + qemu_mutex_lock(&s->mutex); > >> + switch (head.type) { > >> + case VIRTIO_IOMMU_T_ATTACH: > >> + tail.status = virtio_iommu_handle_attach(s, iov, iov_cnt); > >> + break; > >> + case VIRTIO_IOMMU_T_DETACH: > >> + tail.status = virtio_iommu_handle_detach(s, iov, iov_cnt); > >> + break; > >> + case VIRTIO_IOMMU_T_MAP: > >> + tail.status = virtio_iommu_handle_map(s, iov, iov_cnt); > >> + break; > >> + case VIRTIO_IOMMU_T_UNMAP: > >> + tail.status = virtio_iommu_handle_unmap(s, iov, iov_cnt); > >> + break; > >> + default: > >> + tail.status = VIRTIO_IOMMU_S_UNSUPP; > >> + } > >> + qemu_mutex_unlock(&s->mutex); > >> + > >> +out: > >> + sz = iov_from_buf(elem->in_sg, elem->in_num, 0, > >> + &tail, sizeof(tail)); > >> + assert(sz == sizeof(tail)); > >> + > >> + virtqueue_push(vq, elem, sizeof(tail)); > > > > s/tail/head/ (though they are the same size)? > That's unclear to me. Similarly when checking against virtio-net.c, the > element is pushed back to the used ring and len is set to the size of > the status with: > > /* > * Control virtqueue data structures > * > * The control virtqueue expects a header in the first sg entry > * and an ack/status response in the last entry. Data for the > * command goes in between. > */ I was referencing the balloon code when reading the patch, e.g., virtio_balloon_handle_output(). Though after I read more carefully I see that other places are using it as you described. Now I tend to agree with you, because virtqueue_push() who calls virtqueue_unmap_sg() used the len to unmap in_sg[] rather than out_sg[]. So please ignore my previous comment. (then I'm not sure whether the usage in the balloon code was correct now...) > > > >> + virtio_notify(vdev, vq); > >> + g_free(elem); > >> + } > >> +} > > > > [...] > > > >> +static void virtio_iommu_set_features(VirtIODevice *vdev, uint64_t val) > >> +{ > >> + VirtIOIOMMU *dev = VIRTIO_IOMMU(vdev); > >> + > >> + dev->acked_features = val; > >> + trace_virtio_iommu_set_features(dev->acked_features); > >> +} > >> + > >> +static const VMStateDescription vmstate_virtio_iommu_device = { > >> + .name = "virtio-iommu-device", > >> + .unmigratable = 1, > > > > Curious, is there explicit reason to not support migration from the > > first version? :) > The state is made of red black trees, lists. For the former there is no > VMSTATE* ready. I am working on it but I think this should be handled > separately Fair enough. Would you mind to add a similar comment above unmigratable? Thanks, -- Peter Xu
next prev parent reply index Thread overview: 55+ messages / expand[flat|nested] mbox.gz Atom feed top 2019-07-30 17:21 [Qemu-devel] [PATCH for-4.2 v10 00/15] VIRTIO-IOMMU device Eric Auger 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 01/15] update-linux-headers: Import virtio_iommu.h Eric Auger 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 02/15] linux-headers: update against 5.3-rc2 Eric Auger 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 03/15] virtio-iommu: Add skeleton Eric Auger 2019-08-15 13:54 ` Peter Xu 2019-08-29 12:18 ` Auger Eric 2019-08-30 1:26 ` Peter Xu [this message] 2019-08-30 8:12 ` Auger Eric 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 04/15] virtio-iommu: Decode the command payload Eric Auger 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 05/15] virtio-iommu: Add the iommu regions Eric Auger 2019-08-16 4:00 ` Peter Xu 2019-08-29 12:51 ` Auger Eric 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 06/15] virtio-iommu: Endpoint and domains structs and helpers Eric Auger 2019-08-16 4:17 ` Peter Xu 2019-11-04 18:31 ` Jean-Philippe Brucker 2019-11-25 13:14 ` Auger Eric 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 07/15] virtio-iommu: Implement attach/detach command Eric Auger 2019-08-16 4:27 ` Peter Xu 2019-08-29 14:24 ` Auger Eric 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 08/15] virtio-iommu: Implement map/unmap Eric Auger 2019-08-19 8:11 ` Peter Xu 2019-09-03 11:37 ` Auger Eric 2019-09-04 1:44 ` Peter Xu 2019-09-04 4:23 ` Tian, Kevin 2019-09-04 5:37 ` Peter Xu 2019-09-04 5:46 ` Tian, Kevin 2019-09-04 7:54 ` Auger Eric 2019-09-04 8:32 ` Peter Xu 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 09/15] virtio-iommu: Implement translate Eric Auger 2019-08-19 8:24 ` Peter Xu 2019-09-03 11:45 ` Auger Eric 2019-09-04 1:58 ` Peter Xu 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 10/15] virtio-iommu: Implement probe request Eric Auger 2019-08-19 12:08 ` Peter Xu 2019-09-03 12:23 ` Auger Eric 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 11/15] virtio-iommu: Expose the IOAPIC MSI reserved region when relevant Eric Auger 2019-07-30 19:38 ` Michael S. Tsirkin 2019-07-30 23:20 ` Tian, Kevin 2019-07-31 9:05 ` Auger Eric 2019-07-31 19:25 ` Michael S. Tsirkin 2019-07-31 19:44 ` Auger Eric 2019-07-31 23:23 ` Tian, Kevin 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 12/15] virtio-iommu: Implement fault reporting Eric Auger 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 13/15] virtio_iommu: Handle reserved regions in translation process Eric Auger 2019-08-19 12:44 ` Peter Xu 2019-09-01 6:38 ` Michael S. Tsirkin 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 14/15] virtio-iommu-pci: Add virtio iommu pci support Eric Auger 2019-07-30 19:35 ` Michael S. Tsirkin 2019-08-01 12:15 ` Auger Eric 2019-08-01 13:06 ` Michael S. Tsirkin 2019-08-01 13:49 ` Auger Eric 2019-09-01 6:40 ` Michael S. Tsirkin 2019-09-04 14:19 ` Auger Eric 2019-09-04 21:36 ` Michael S. Tsirkin 2019-07-30 17:21 ` [Qemu-devel] [PATCH for-4.2 v10 15/15] hw/arm/virt: Add the virtio-iommu device tree mappings Eric Auger
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=20190830012614.GK8729@xz-x1 \ --to=peterx@redhat.com \ --cc=alex.williamson@redhat.com \ --cc=bharat.bhushan@nxp.com \ --cc=eric.auger.pro@gmail.com \ --cc=eric.auger@redhat.com \ --cc=jean-philippe@linaro.org \ --cc=kevin.tian@intel.com \ --cc=mst@redhat.com \ --cc=peter.maydell@linaro.org \ --cc=qemu-arm@nongnu.org \ --cc=qemu-devel@nongnu.org \ --cc=tn@semihalf.com \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: link
QEMU-Devel Archive on lore.kernel.org Archives are clonable: git clone --mirror https://lore.kernel.org/qemu-devel/0 qemu-devel/git/0.git git clone --mirror https://lore.kernel.org/qemu-devel/1 qemu-devel/git/1.git git clone --mirror https://lore.kernel.org/qemu-devel/2 qemu-devel/git/2.git # If you have public-inbox 1.1+ installed, you may # initialize and index your mirror using the following commands: public-inbox-init -V2 qemu-devel qemu-devel/ https://lore.kernel.org/qemu-devel \ qemu-devel@nongnu.org public-inbox-index qemu-devel Example config snippet for mirrors Newsgroup available over NNTP: nntp://nntp.lore.kernel.org/org.nongnu.qemu-devel AGPL code for this site: git clone https://public-inbox.org/public-inbox.git