[dpdk-dev] [PATCH 2/3] app/testpmd: enabled control for packet timestamps

[Olivier Matz]

Hi,

On Fri, 28 Apr 2017 11:12:29 +0200, Thomas Monjalon <thomas at monjalon.net> wrote:
> 28/04/2017 02:19, Wu, Jingjing:
> > > 25/04/2017 16:02, Wu, Jingjing:  
> > > > From: Oleg Kuporosov  
> > > > > Implemented two methods of control
> > > > >
> > > > > - by --enable-timestamps CL testpmd application we can enable  
> > > timestamping  
> > > > >   for all ports;
> > > > > - in interactive mode port config <port> timestamps on|off is able to
> > > > >   configure timestamping per specific port.
> > > > >
> > > > > The control doesn't interact with IEEE1588 PTP implementation there
> > > > > as it is under macro compilation but can be extended in the future.
> > > > >
> > > > > This feature is required for debugging/testing purposes for real
> > > > > time HW packet timestamping.  
> > > >
> > > > We have ieee1588fwd.c to demo the timesync enable/disable, can we
> > > > reuse The fwd engine instead of defining new commands?  
> > > 
> > > Yes for IEEE1588 feature, we should use app/test-pmd/ieee1588fwd.c.
> > > 
> > > There is more to say about this feature.
> > > 
> > > The main goal of this patchset was to add a timestamp in the mbuf.
> > > It has been done by another patchset in 17.05.  
> > 
> > OK. But it is not clear now what is the timestamp for, right?  
> 
> Olivier can comment on the Rx timestamp (PKT_RX_TIMESTAMP).

The timestamp can be added in a mbuf by a PMD or by software.
The unit and time reference (the 0) is not normalized: they
can be different between 2 PMDs but must be the same for one
PMD on the same port.

As of now, this field could be used by an application to know when
a packet was effectively received by the hardware. The use case could
be high frequency trading for instance. The PMD can provide an API to
convert timestamp value into another unit.

Possible evolutions are:
- a library to normalize and compare timestamps from different
  sources
- implementations in more PMDs



> 
> Please Jingjing (or Pablo or Daniel),
> could you explain the relation with PKT_TX_IEEE1588_TMST?
> 
> > > Do we know how to test this timestamp in testpmd?  
> > 
> > Mbuf dump can show this value. The problem is if we can use the
> > rte_eth_timesync_enable/disable to indicate the timestamp
> > is in mbuf or not.  
> 
> Please could you describe the problem?
> 
> > > About IEEE1588 feature, why is there a config option?
> > > 	CONFIG_RTE_LIBRTE_IEEE1588
> > > A feature should never be disabled at compile time.
> > > There is also a runtime enablement with rte_eth_timesync_enable().
> > > 
> > > I think we need some discussions here.  
> > 
> > Yes, I agree.  
> 
> What is your opinion?
> What prevents from removing CONFIG_RTE_LIBRTE_IEEE1588?
>