Re: [virtio-dev] Re: [virtio-comment] Re: [virtio-dev] RE: [virtio-comment] RE: [PATCH v19] virtio-net: support inner header hash

["Michael S. Tsirkin" <mst@redhat.com>]

On Fri, Jun 30, 2023 at 02:15:13PM +0800, Heng Qi wrote:
> 
> 
> å 2023/6/30 äå1:59, Michael S. Tsirkin åé:
> > On Fri, Jun 30, 2023 at 09:55:41AM +0800, Heng Qi wrote:
> > > 
> > > å 2023/6/30 äå9:36, Parav Pandit åé:
> > > > > From: Heng Qi <hengqi@linux.alibaba.com>
> > > > > Sent: Thursday, June 29, 2023 8:54 PM
> > > > > 
> > > > > On Thu, Jun 29, 2023 at 04:59:28PM +0000, Parav Pandit wrote:
> > > > > > > From: Michael S. Tsirkin <mst@redhat.com>
> > > > > > > Sent: Thursday, June 29, 2023 7:48 AM
> > > > > > > > > struct virtio_net_hash_config reserved is fine.
> > > > > > > > +1.
> > > > > > > > 
> > > > > > > > Inner header hash is orthogonal to RSS, and it's fine to have its
> > > > > > > > own structure and commands.
> > > > > > > > There is no need to send additional RSS fields when we configure
> > > > > > > > inner header hash.
> > > > > > > > 
> > > > > > > > Thanks.
> > > > > > > Not RSS, hash calculations. It's not critical, but I note that
> > > > > > > practically you said you will enable this with symmetric hash so it
> > > > > > > makes sense to me to send this in the same command with the key.
> > > > > > > 
> > > > > > In the v19, we have,
> > > > > > 
> > > > > > +\item[VIRTIO_NET_F_HASH_TUNNEL] Requires VIRTIO_NET_F_CTRL_VQ
> > > > > along with VIRTIO_NET_F_RSS or VIRTIO_NET_F_HASH_REPORT.
> > > > > > So it is done along with rss, so in same struct as rss config is fine.
> > > > > Do you mean having both virtio_net_rss_config and virtio_net_hash_config
> > > > > have enabled_hash_types?
> > > > > Like this:
> > > > > 
> > > > > struct virtio_net_rss_config {
> > > > >        le32 hash_types;
> > > > >        le16 indirection_table_mask;
> > > > >        struct rss_rq_id unclassified_queue;
> > > > >        struct rss_rq_id indirection_table[indirection_table_length];
> > > > >        le16 max_tx_vq;
> > > > >        u8 hash_key_length;
> > > > >        u8 hash_key_data[hash_key_length];
> > > > > +    le32 enabled_tunnel_types;
> > > > > };
> > > > > 
> > > > > struct virtio_net_hash_config {
> > > > >        le32 hash_types;
> > > > > -    le16 reserved[4];
> > > > > +    le32 enabled_tunnel_types;
> > > > > +    le16 reserved[2];
> > > > >        u8 hash_key_length;
> > > > >        u8 hash_key_data[hash_key_length];
> > > > > };
> > Oh, I forgot that rss and hash had identical structures.
> > And we want to keep that I think.
> > 
> > But now it's not clear to me: does the same enabled_tunnel_types
> > apply to both hash calculation and rss?
> 
> Yes. What I'm trying to say is that making the inner header hash and
> RSS/hash calculation clear delimitation will make everything easier.
> The inner header hash just configures enabled_tunnel_types.
> The RSS/hash calculation is configured as before without modification.
> 
> > I note we normally have separate configs for hash and rss.
> > So to keep it consistent what should we do? two set commands?
> 
> As I explained above, like outer udp port hash/symmetric toeplitz hash,
> these fall under the umbrella of RSS/hash calculation.

Yes but note that symmetric toeplitz hash has to be configured
separately for RSS and for hashing.

> Let's keep the inner header hash simple.
> 
> > Or does enabled_tunnel_types apply to both rss and hash?
> 
> Certainly. See:
> 
> ÂÂÂ +\item[VIRTIO_NET_F_HASH_TUNNEL] Requires VIRTIO_NET_F_CTRL_VQ along
> with VIRTIO_NET_F_RSS or VIRTIO_NET_F_HASH_REPORT.


It does not really say that.



> > 
> > We should have reserved more space. We can still do it if you like:
> > 
> > struct virtio_net_rss_tunnel_config {
> >        le32 enabled_tunnel_types;
> >        le16 reserved[6];
> >        struct virtio_net_rss_config hash;
> > };
> > 
> > struct virtio_net_hash_tunnel_config {
> >        le32 enabled_tunnel_types;
> >        le16 reserved[6];
> >        struct virtio_net_hash_config hash;
> > };
> > 
> > ?
> > 
> > 
> > 
> > 
> > 
> > > > > 
> > > > > If yes, this should have been discussed in v10 [1] before, enabled_tunnel_types
> > > > > in virtio_net_rss_config will follow the variable length field and cause
> > > > > misalignment.
> > > > > 
> > > > > If we let the inner header hash reuse the virtio_net_hash_config structure, it
> > > > > can work, but the only disadvantage is that the configuration of the inner
> > > > > header hash and *RSS*(not hash calculations) becomes somewhat coupled.
> > > > > Just imagine:
> > > > > If the driver and the device negotiated VIRTIO_NET_F_HASH_TUNNEL and
> > > > > VIRTIO_NET_F_RSS, but did not negotiate VIRTIO_NET_F_HASH_REPORT, 1.
> > > > > then if we only want to configure the inner header hash (such as
> > > > > enabled_tunnel_types), it is good for us to send virtio_net_hash_config alone;
> > > > > 2. but then if we want to configure the inner header hash and RSS (such as
> > > > > indirection table), we need to send all virtio_net_rss_config and
> > > > > virtio_net_hash_config once, because virtio_net_rss_config now does not carry
> > > > > enabled_tunnel_types due to misalignment.
> > > > > 
> > > > > So, I think the following structure will make it clearer to configure inner header
> > > > > hash and RSS/hash calculation.
> > > > > But in any case, if we still propose to reuse virtio_net_hash_config proposal, I
> > > > > am ok, no objection:
> > > > > 
> > > > > 1. The supported_tunnel_types are placed in the device config space;
> > > > > 
> > > > Yes. I forgot the variable length part.
> > > > The second disadvantage I remember now is one need to resupply all the rss hash config parameter and device needs to compare and modify for this one field.
> > Or it could be an advantage since one normally wants to configure a
> > symmetric key with this. Further device can just use the new config
> 
> When we want to configure the hash key, he continues to use the previous
> rss/hash calculation interface. This is ok.
> 
> Thanks.

I don't understand this sentence. My point is simply that
to use the tunnel key has to be symmetric. So two commands
will be required: one to set tunnel types, one to
set the key.


> > with no need to check what the old one was. I'd call it a wash.
> > 
> > > > Given these two disadvantages, I also prefer independent SET command the way you have it.
> > > OK, let's wait for Michael's input again.
> > > 
> > > Thanks.
> > 
> > This part is not critical to me, but now I understand we need two sets of SET commands.
> > 
> > 
> > > > > 2.
> > > > > Reserve the following structure:
> > > > > 
> > > > >         struct virtnet_hash_tunnel {
> > > > > le32 enabled_tunnel_types;
> > > > >         };
> > > > > 
> > > > > 3. Reserve the SET command for enabled_tunnel_types and remove the GET
> > > > > command for enabled_tunnel_types.
> > > > > 
> > > > > [1] https://lists.oasis-open.org/archives/virtio-dev/202303/msg00317.html
> > > > > 
> > > > > Thanks a lot!
> > 
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: virtio-dev-unsubscribe@lists.oasis-open.org
> > For additional commands, e-mail: virtio-dev-help@lists.oasis-open.org