[dpdk-dev] [PATCH v3 0/3] enhance TX checksum command and csum forwarding engine

Ananyev, Konstantin konstantin.ananyev at intel.com
Thu Jan 8 11:54:41 CET 2015


Hi Frank,

> -----Original Message-----
> From: Liu, Jijiang
> Sent: Thursday, January 08, 2015 8:52 AM
> To: Ananyev, Konstantin; 'Olivier MATZ'
> Cc: dev at dpdk.org
> Subject: RE: [dpdk-dev] [PATCH v3 0/3] enhance TX checksum command and csum forwarding engine
> 
> Hi,
> 
> > -----Original Message-----
> > From: Ananyev, Konstantin
> > Sent: Wednesday, January 7, 2015 8:07 PM
> > To: Liu, Jijiang; 'Olivier MATZ'
> > Cc: dev at dpdk.org
> > Subject: RE: [dpdk-dev] [PATCH v3 0/3] enhance TX checksum command and
> > csum forwarding engine
> >
> >
> >
> > > -----Original Message-----
> > > From: Liu, Jijiang
> > > Sent: Wednesday, January 07, 2015 11:39 AM
> > > To: Ananyev, Konstantin; 'Olivier MATZ'
> > > Cc: dev at dpdk.org
> > > Subject: RE: [dpdk-dev] [PATCH v3 0/3] enhance TX checksum command and
> > > csum forwarding engine
> > >
> > > Hi Konstantin,
> > >
> > > > -----Original Message-----
> > > > From: Ananyev, Konstantin
> > > > Sent: Wednesday, January 7, 2015 5:59 PM
> > > > To: Liu, Jijiang; 'Olivier MATZ'
> > > > Cc: dev at dpdk.org
> > > > Subject: RE: [dpdk-dev] [PATCH v3 0/3] enhance TX checksum command
> > > > and csum forwarding engine
> > > >
> > > > Hi Frank,
> > > >
> > > > > -----Original Message-----
> > > > > From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Liu, Jijiang
> > > > > Sent: Wednesday, January 07, 2015 2:04 AM
> > > > > To: 'Olivier MATZ'
> > > > > Cc: dev at dpdk.org
> > > > > Subject: Re: [dpdk-dev] [PATCH v3 0/3] enhance TX checksum command
> > > > > and csum forwarding engine
> > > > >
> > > > > Hi Olivier,
> > > > >
> > > > > > -----Original Message-----
> > > > > > From: Olivier MATZ [mailto:olivier.matz at 6wind.com]
> > > > > > Sent: Saturday, December 13, 2014 12:33 AM
> > > > > > To: Liu, Jijiang
> > > > > > Cc: dev at dpdk.org
> > > > > > Subject: Re: [dpdk-dev] [PATCH v3 0/3] enhance TX checksum
> > > > > > command and csum forwarding engine
> > > > > >
> > > > > > Hello,
> > > > > >
> > > > > > On 12/12/2014 04:48 AM, Liu, Jijiang wrote:
> > > > > > > The 'hw/sw' option  is used to set/clear the flag of enabling
> > > > > > > TX tunneling packet
> > > > > > checksum hardware offload in testpmd application.
> > > > > >
> > > > > > This is not clear at all.
> > > > > > In your command, there is (hw|sw|none).
> > > > > > Are you talking about inner or outer?
> > > > > > Is this command useful for any kind of packet?
> > > > > > How does it combine with "tx_checksum set outer-ip (hw|sw)"?
> > > > > >
> > > > >
> > > > > I rethink these TX checksum commands in this patch set and agree
> > > > > with you that we should make some changes for having clear meaning for
> > them.
> > > > >
> > > > > There are  3 commands in patch set as follows, 1. tx_checksum set
> > > > > tunnel (hw|sw|none) (port-id)
> > > > >
> > > > > Now I also think the command 1 may confuse user, they probably
> > > > > don't understand  why we need 'hw' or 'sw' option and when  to use
> > > > > the two option, so I will replace the command with 'tx_checksum
> > > > > set hw-tunnel-mode
> > > > (on|off) (port-id)' command.
> > > >
> > > > I am a bit confused here, could you explain what would be a
> > > > behaviour for 'on' and 'off'?
> > > > Konstantin
> > >
> > > I have explained the behaviour for 'on' and'off' below,
> > >
> > > The command 'tx_checksum set hw-tunnel-mode (on|off)  (port-id)' is
> > > used to set/clear  TESTPMD_TX_OFFLOAD_TUNNEL_CKSUM flag.
> > >
> > > Actually, the PKT_TX_UDP_TUNNEL_PKT offload flag will be set if the
> > > testpmd flag is set, which means to tell HW treat  that transmit
> > > packet as a tunneling packet to do checksum offload  When 'on' is set, which is
> > able to meet Method B.1 and method C.
> > >
> > > When 'off' is set, the TESTPMD_TX_OFFLOAD_TUNNEL_CKSUM is not needed
> > > to set, so the PKT_TX_UDP_TUNNEL_PKT offload flag is not needed to
> > > set,  then HW treat  that transmit packet as a non-tunneling packet. It is able to
> > meet Method B.2.
> > >
> > > Is the explanation not clear?
> >
> > Ok, and how I can set method A (testpmd treat all packets as non-tunnelling and
> > never look beyond outer L4 header) then?
> > Konstantin
> 
> 
> > > > > As to case A, I think it is not mandatory to cover it in csum fwd
> > > > > engine for tunneling packet.
> 
> If you think the case A is essential, and it must be covered in  csum fwd, then we can add a command:
> 
> tx_checksum set sw-tunnel-mode (on|off)  (port-id)
> 
> if the 'off' is set ,  csum fwd engine don't check  if that packet is a tunneling packet and treat all packets as non-tunneling and never
> look beyond outer L4 header.
> 
> if the 'on' is set,  csum fwd engine will check if that packet is a tunneling packet.
> 
> And we are able to test all of cases in  http://dpdk.org/ml/archives/dev/2014-December/009213.html
> 
> Test case A:
> 
> tx_checksum set sw-tunnel-mode off
> tx_checksum set hw-tunnel-mode off
> tx_checksum set  ip   hw
> 
> test case B.1:
> 
> tx_checksum set sw-tunnel-mode on
> tx_checksum set hw-tunnel-mode on
> tx_checksum set  ip   hw
> tx_checksum set  tcp   hw
> 
> test case B.2:
> 
> tx_checksum set sw-tunnel-mode on
> tx_checksum set hw-tunnel-mode off
> tx_checksum set  ip   hw
> 
> test case C:
> 
> tx_checksum set sw-tunnel-mode on
> tx_checksum set hw-tunnel-mode on
> tx_checksum set  outer-ip   hw
> tx_checksum set  ip   hw
> tx_checksum set  tcp   hw
> 
> 
>  In addition, the reason of discarding ' tx_checksum set  tunnel (hw|sw|none) (port-id)' command is that  user probably confuse the
> following case.
>  tx_checksum set  tunnel sw
> tx_checksum set  ip   hw
> 
> In fact, we are still using hardware TX checksum offload in this case,  but the command " tx_checksum set  tunnel sw"  seems tell user
> that software compute the checksum.

So  ' set hw-tunnel-mode on' implies that ' sw-tunnel-mode' is already set to 'on'?
What would happen if user would do:
tx_checksum set sw-tunnel-mode off
tx_checksum set hw-tunnel-mode on
Or this combination is not allowed? 

As long as we have all 4 cases covered, I don't really care would it be:
tx_checksum set  tunnel (hw|sw|none)
or 
tx_checksum set sw-tunnel-mode   (on|off)
tx_checksum set hw-tunnel-mode  (on|off)

Though having 2 commands instead of 1 seems like overcomplicating and  more error prone...
Might be:
tx_checksum set tunnel-mode (none | hw | sw)
?

Konstantin



> 
> > >
> > > >
> > > > >
> > > > > 2. tx_checksum set outer-ip (hw|sw) (port-id) 3. tx_checksum set
> > > > > (ip|udp|tcp|sctp) (hw|sw) (port-id)
> > > > >
> > > > > The command 2 will be merged into command 3, the new command is '
> > > > > tx_checksum set  (outer-ip|ip|udp|tcp|sctp) (hw|sw) (port- id)'.
> > > > >
> > > > > These most of the cases in
> > > > > http://dpdk.org/ml/archives/dev/2014-December/009213.html will be
> > > > > covered by using the two commands
> > > > >
> > > > > The command 'tx_checksum set hw-tunnel-mode (on|off)  (port-id)'
> > > > > is used to set/clear  TESTPMD_TX_OFFLOAD_TUNNEL_CKSUM flag.
> > > > > Actually, the PKT_TX_UDP_TUNNEL_PKT offload flag will be set if
> > > > > the testpmd flag is set, which tell driver/HW treat  that transmit
> > > > > packet as a
> > > > tunneling packet.
> > > > >
> > > > > When 'on' is set, which is able to meet Method B.1 and method C.
> > > > >
> > > > > When 'off' is set, the TESTPMD_TX_OFFLOAD_TUNNEL_CKSUM is not
> > > > > needed to set, so the PKT_TX_UDP_TUNNEL_PKT offload flag is not
> > > > > needed to set,  then
> > > > HW treat  that transmit packet as a non-tunneling packet. It is able
> > > > to meet Method B.2.
> > > > >
> > > > > As to case A, I think it is not mandatory to cover it in csum fwd
> > > > > engine for
> > > > tunneling packet.
> > > > >
> > > > > Is the above description clear for you?
> > > > >
> > > > > > Regards,
> > > > > > Olivier



More information about the dev mailing list