Re: [PATCH ethtool-next] rxclass: Make output for RSS context action explicit

[Joe Damato <jdamato@fastly.com>]

On Fri, Nov 08, 2024 at 12:34:49PM -0800, Joe Damato wrote:
> On Fri, Nov 08, 2024 at 07:56:41PM +0000, Edward Cree wrote:
> > On 08/11/2024 19:32, Daniel Xu wrote:
> > > Currently, if the action for an ntuple rule is to redirect to an RSS
> > > context, the RSS context is printed as an attribute. At the same time,
> > > a wrong action is printed. For example:
> > > 
> > >     # ethtool -X eth0 hfunc toeplitz context new start 24 equal 8
> > >     New RSS context is 1
> > > 
> > >     # ethtool -N eth0 flow-type ip6 dst-ip $IP6 context 1
> > >     Added rule with ID 0
> > > 
> > >     # ethtool -n eth0 rule 0
> > >     Filter: 0
> > >             Rule Type: Raw IPv6
> > >             Src IP addr: :: mask: ffff:ffff:ffff:ffff:ffff:ffff:ffff:ffff
> > >             Dest IP addr: <redacted> mask: ::
> > >             Traffic Class: 0x0 mask: 0xff
> > >             Protocol: 0 mask: 0xff
> > >             L4 bytes: 0x0 mask: 0xffffffff
> > >             RSS Context ID: 1
> > >             Action: Direct to queue 0
> > > 
> > > This is wrong and misleading. Fix by treating RSS context as a explicit
> > > action. The new output looks like this:
> > > 
> > >     # ./ethtool -n eth0 rule 0
> > >     Filter: 0
> > >             Rule Type: Raw IPv6
> > >             Src IP addr: :: mask: ffff:ffff:ffff:ffff:ffff:ffff:ffff:ffff
> > >             Dest IP addr: <redacted> mask: ::
> > >             Traffic Class: 0x0 mask: 0xff
> > >             Protocol: 0 mask: 0xff
> > >             L4 bytes: 0x0 mask: 0xffffffff
> > >             Action: Direct to RSS context id 1
> > > 
> > > Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
> > 
> > I believe this patch is incorrect.  My understanding is that on
> >  packet reception, the integer returned from the RSS indirection
> >  table is *added* to the queue number from the ntuple rule, so
> >  that for instance the same indirection table can be used for one
> >  rule distributing packets over queues 0-3 and for another rule
> >  distributing a different subset of packets over queues 4-7.
> > I'm not sure if this behaviour is documented anywhere, and
> >  different NICs may have different interpretations, but this is
> >  how sfc ef10 behaves.
> 
> I just wanted to chime in and say that my understanding has always
> been more aligned with Daniel's and I had also found the ethtool
> output confusing when directing flows that match a rule to a custom
> context.
> 
> If Daniel's patch is wrong (I don't know enough to say if it is or
> not), would it be possible to have some alternate ethtool output
> that's less confusing? Or for this specific output to be outlined in
> the documentation somewhere?

Sorry for the quick follow up, but I just tested this and I do think
there is an issue with ethtool's output.

Here's an example from my system where I create 18 queues and a
custom RSS context to send flows to queues 16 and 17 only:

$ ethtool --version
ethtool version 6.7

$ sudo ethtool -L eth2 combined 18

$ sudo ethtool -X eth2 weight 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 context new
New RSS context is 1

$ sudo ethtool -U eth2 flow-type tcp4 dst-port 11211 context 1
Added rule with ID 1023

$ sudo ethtool -n eth2 rule 1023
Filter: 1023
	Rule Type: TCP over IPv4
	Src IP addr: 0.0.0.0 mask: 255.255.255.255
	Dest IP addr: 0.0.0.0 mask: 255.255.255.255
	TOS: 0x0 mask: 0xff
	Src port: 0 mask: 0xffff
	Dest port: 11211 mask: 0x0
	RSS Context ID: 1
	Action: Direct to queue 0

I don't understand why this would say "Direct to queue 0" when the
weights specifically disallow this for context 1.