Re: [PATCH v3 01/17] virtio_pci: struct virtio_pci_common_cfg add queue_notify_data

[Jason Wang <jasowang@redhat.com>]

On Tue, Feb 8, 2022 at 11:29 AM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote:
>
> On Tue, 8 Feb 2022 11:24:13 +0800, Jason Wang <jasowang@redhat.com> wrote:
> > On Tue, Feb 8, 2022 at 11:20 AM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote:
> > >
> > > On Tue, 8 Feb 2022 11:03:17 +0800, Jason Wang <jasowang@redhat.com> wrote:
> > > > On Tue, Feb 8, 2022 at 10:17 AM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote:
> > > > >
> > > > > On Mon, 7 Feb 2022 16:06:15 +0800, Jason Wang <jasowang@redhat.com> wrote:
> > > > > > On Mon, Feb 7, 2022 at 2:07 PM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote:
> > > > > > >
> > > > > > > On Mon, 7 Feb 2022 11:41:06 +0800, Jason Wang <jasowang@redhat.com> wrote:
> > > > > > > >
> > > > > > > > 在 2022/1/26 下午3:35, Xuan Zhuo 写道:
> > > > > > > > > Add queue_notify_data in struct virtio_pci_common_cfg, which comes from
> > > > > > > > > here https://github.com/oasis-tcs/virtio-spec/issues/89
> > > > > > > > >
> > > > > > > > > Since I want to add queue_reset after it, I submitted this patch first.
> > > > > > > > >
> > > > > > > > > Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
> > > > > > > > > ---
> > > > > > > > >   include/uapi/linux/virtio_pci.h | 1 +
> > > > > > > > >   1 file changed, 1 insertion(+)
> > > > > > > > >
> > > > > > > > > diff --git a/include/uapi/linux/virtio_pci.h b/include/uapi/linux/virtio_pci.h
> > > > > > > > > index 3a86f36d7e3d..492c89f56c6a 100644
> > > > > > > > > --- a/include/uapi/linux/virtio_pci.h
> > > > > > > > > +++ b/include/uapi/linux/virtio_pci.h
> > > > > > > > > @@ -164,6 +164,7 @@ struct virtio_pci_common_cfg {
> > > > > > > > >     __le32 queue_avail_hi;          /* read-write */
> > > > > > > > >     __le32 queue_used_lo;           /* read-write */
> > > > > > > > >     __le32 queue_used_hi;           /* read-write */
> > > > > > > > > +   __le16 queue_notify_data;       /* read-write */
> > > > > > > > >   };
> > > > > > > >
> > > > > > > >
> > > > > > > > So I had the same concern as previous version.
> > > > > > > >
> > > > > > > > This breaks uABI where program may try to use sizeof(struct
> > > > > > > > virtio_pci_common_cfg).
> > > > > > > >
> > > > > > > > We probably need a container structure here.
> > > > > > >
> > > > > > > I see, I plan to add a struct like this, do you think it's appropriate?
> > > > > > >
> > > > > > > struct virtio_pci_common_cfg_v1 {
> > > > > > >         struct virtio_pci_common_cfg cfg;
> > > > > > >         __le16 queue_notify_data;       /* read-write */
> > > > > > > }
> > > > > >
> > > > > > Something like this but we probably need a better name.
> > > > >
> > > > >
> > > > > how about this?
> > > > >
> > > > >         /* Ext Fields in VIRTIO_PCI_CAP_COMMON_CFG: */
> > > > >         struct virtio_pci_common_cfg_ext {
> > > > >                 struct virtio_pci_common_cfg cfg;
> > > > >
> > > > >                 __le16 queue_notify_data;       /* read-write */
> > > > >
> > > > >                 __le16 reserved0;
> > > > >                 __le16 reserved1;
> > > > >                 __le16 reserved2;
> > > > >                 __le16 reserved3;
> > > > >                 __le16 reserved4;
> > > > >                 __le16 reserved5;
> > > > >                 __le16 reserved6;
> > > > >                 __le16 reserved7;
> > > > >                 __le16 reserved8;
> > > > >                 __le16 reserved9;
> > > > >                 __le16 reserved10;
> > > > >                 __le16 reserved11;
> > > > >                 __le16 reserved12;
> > > > >                 __le16 reserved13;
> > > > >                 __le16 reserved14;
> > > > >         };
> > > >
> > > > I still think the container without padding is better. Otherwise
> > > > userspace needs to use offset_of() trick instead of sizeof().
> > >
> > > In this case, as virtio_pci_common_cfg_ext adds new members in the future, we
> > > will add more container structures.
> > >
> > > In that case, I think virtio_pci_common_cfg_v1 is a good name instead.
> >
> > Something like "virtio_pci_common_cfg_notify" might be a little bit better.
>
> Although there is only one notify_data in this patch, I plan to look like this
> after my patch set:
>
>         struct virtio_pci_common_cfg_v1 {
>                 struct virtio_pci_common_cfg cfg;
>
>                 __le16 queue_notify_data;       /* read-write */
>                 __le16 queue_reset;       /* read-write */
>         }
>
> If we use virtio_pci_common_cfg_notify, then we will get two structures after
> this patch set:
>
>         struct virtio_pci_common_cfg_notify {
>                 struct virtio_pci_common_cfg cfg;
>
>                 __le16 queue_notify_data;       /* read-write */
>         }
>
>         struct virtio_pci_common_cfg_reset {
>                 struct virtio_pci_common_cfg_notify cfg;
>
>                 __le16 queue_reset;       /* read-write */
>         }

Right, this is sub-optimal, and we need padding in cfg_notify
probably. But I couldn't think of a better idea currently, maybe we
can listen from others opinion

But we use something like this for vnet_header extension

struct virtio_net_hdr_v1{
};

struct virtio_net_hdr_v1_hash{
struct virtio_net_hdr_v1;
__le32 XXX;
...
__le16 padding;
};

And it's not hard to imagine there would be another container for
struct virtio_net_hdr_v1_hash in the future if we want to extend vnet
header.

Thanks

>
>
> Thanks.
>
> >
> > Thanks
> >
> > >
> > > Thanks.
> > >
> > >
> > > >
> > > > Thanks
> > > >
> > > > >
> > > > > Thanks
> > > > >
> > > > > >
> > > > > > Thanks
> > > > > >
> > > > > > >
> > > > > > > Thanks.
> > > > > > >
> > > > > > > >
> > > > > > > > THanks
> > > > > > > >
> > > > > > > >
> > > > > > > > >
> > > > > > > > >   /* Fields in VIRTIO_PCI_CAP_PCI_CFG: */
> > > > > > > >
> > > > > > >
> > > > > >
> > > > >
> > > >
> > >
> >
>