[dpdk-dev] [PATCH] net/ixgbe: put 5tuple check in front to jump over ntuple filter case

Zhang, Qi Z qi.z.zhang at intel.com
Mon Oct 15 05:30:44 CEST 2018


Hi Wei:

> -----Original Message-----
> From: Zhao1, Wei
> Sent: Thursday, October 11, 2018 1:10 AM
> To: Zhang, Qi Z <qi.z.zhang at intel.com>; mocan <faicker.mo at ucloud.cn>
> Cc: dev at dpdk.org; Lu, Wenzhuo <wenzhuo.lu at intel.com>
> Subject: RE: Re:RE: [dpdk-dev] [PATCH] net/ixgbe: put 5tuple check in front
> to jump over ntuple filter case
> 
> Hi, qi
> 
> > -----Original Message-----
> > From: Zhang, Qi Z
> > Sent: Thursday, October 11, 2018 2:36 AM
> > To: Zhao1, Wei <wei.zhao1 at intel.com>; mocan <faicker.mo at ucloud.cn>
> > Cc: dev at dpdk.org; Lu, Wenzhuo <wenzhuo.lu at intel.com>
> > Subject: RE: Re:RE: [dpdk-dev] [PATCH] net/ixgbe: put 5tuple check in
> > front to jump over ntuple filter case
> >
> >
> >
> > > -----Original Message-----
> > > From: Zhao1, Wei
> > > Sent: Monday, October 8, 2018 2:46 AM
> > > To: Zhang, Qi Z <qi.z.zhang at intel.com>; mocan <faicker.mo at ucloud.cn>
> > > Cc: dev at dpdk.org; Lu, Wenzhuo <wenzhuo.lu at intel.com>
> > > Subject: RE: Re:RE: [dpdk-dev] [PATCH] net/ixgbe: put 5tuple check
> > > in front to jump over ntuple filter case
> > >
> > > Hi,
> > >
> > >
> > > > -----Original Message-----
> > > > From: Zhang, Qi Z
> > > > Sent: Wednesday, September 26, 2018 7:14 PM
> > > > To: mocan <faicker.mo at ucloud.cn>; Zhao1, Wei
> <wei.zhao1 at intel.com>
> > > > Cc: dev at dpdk.org
> > > > Subject: RE: Re:RE: [dpdk-dev] [PATCH] net/ixgbe: put 5tuple check
> > > > in front to jump over ntuple filter case
> > > >
> > > > OK, got your point. We should not reject a possible valid fdir
> > > > flow at n-tuple flow check stage.
> > > >
> > > > Review-by: Qi Zhang <qi.z.zhang at intel.com>
> > >
> > >
> > > I agree with the point of " We should not reject a possible valid
> > > fdir flow at n-tuple flow check stage".
> > > But, I think the fix patch should be more generic for all types
> > > filter of this problem.
> > > Maybe, we should delete all " goto out"  in function
> ixgbe_flow_create().
> > > Then, it will go to ntuple filter and  ethertype filter, syn filter
> > > and fdir filter ,l2_tn_filter one by one.
> > > And aslo, we should code as
> > >
> > > {
> > >
> > > Ntuple:
> > > ..........
> > > if(ret)
> > >     Goto ethertype
> > > ..........
> > >
> > > Ethertype:
> > >
> > > ..........
> > > if(ret)
> > >     Goto fdir filter
> > > .........
> > >
> > > fdir filter:
> > >
> > > if(ret)
> > >   Goto l2_tn_filter
> > >
> > > l2_tn_filter:
> > >
> > > .............
> > >
> > > }
> > >
> > > This fix patch only solve the problem of  ntuple and fdir.
> > > Qi, What do you think of this?
> >
> > I'm not the author of this part of code, so my understanding of
> > current implementation is:
> > It assume a flow will not be ambiguous which means if it match to some
> > filter parser (ixgbe_parse_xxx_filter), it is not necessary to match
> > on a different filter.
> > But I'm not sure if the assumption is correct or not, (this depends on
> > the knowledge of the device capability), So do you mean the assumption
> > is not correct? If you think a generic fix is necessary, I have below
> > comments
> 
> Yes, the assumption is may cause bug, this patch is an evidence, maybe this
> user has encountered this problem.
> 
> >
> > 1. it is better be done by Intel people with enough validation
> 
> I agree with you, I will commit a generic fix patch later.
> 
> >2. two options  for patch submit.
> > 	Submit a v2 with the generic fix, and it will be captured in this release.
> > 	If it is not urgent, we can just accept current one first, then have
> >a  separate patch in next release.
> 
> Ok, If someone supply a v2 with the generic fix, I will ack.
> 

Just want to confirm with you , are you agree to merge this patch?
Or you think v2 with generic fix is necessary?
>From my view, the patch can be accepted, since it just add more strict check in cons_parse_ntuple_filter, it does not break anything, and it fix the specific issue.

Thanks
Qi





More information about the dev mailing list