RE: [PATCH net-next v3 08/14] net: ethernet: oa_tc6: Remove FCS size in RX frame

[Selvamani Rajagopal]

> > > Subject: Re: [PATCH net-next v3 08/14] net: ethernet: oa_tc6: Remove FCS size in
> >
> > My understanding is that in TX side, MAC will pad FCS automatically.
> 
> TX pad is separate from the FCS.
> The frames are padded to a minimum size so that collision detect works
> on maximal length 10M coax segments.
> (Or rather coax+link+coax+link+coax acting as a single collision domain.)
> The same minimum frame length is used on modern networks which are all
> physically point-to-point and all 'hubs' are store+forward where a minimum
> packet length isn't strictly needed.
> 
> VLAN headers can give unexpected amounts of padding - VLAN didn't exist
> when I was writing ethernet drivers!

Understood and agree. Mentiong "pad FCS" may be the cause for the confusion.
MAC (at least our MAC) does PAD + FCS.

> 
> 
> Is this an SPI interface to the ethernet MAC?

Yes. Correct. Twisted pair Ethernet on one end and SPI interface on the other end.

> You might want to pass the FCS over the SPI link and then verify in software
> before treating the frame as valid, deleting the FCS, and passing the frame to
> the protocol stack.
> This would give extra protection for errors on the SPI link that might not
> otherwise be detected.
> (Although if you distrust the SPI link you'd what to pass the TX FCS as well.)
> 
> Unless you are doing IP checksum offload the IP checksum will pick up TCP/UDP errors.
> 
> But if you are running ISO transport (which we did 30 years ago) there is
> no other checksum. So if/when you have a VMEbus card that suffers ground


Our MAC does give Ethernet frame with FCS to the SPI host. That's why the modification I did to oa_tc6.c,
subtracts FCS size before giving to stack. But I didn't add FCS sanity check at the host after assembling Ethernet
frame from SPI data chunks. Doing so would take good CPU time and performance may suffer. 

At least, OA TC6 framework didn't implement that (yet)


> bounce when, for example, passing a repeated 0xff, 0xff, 0x00 sequence you
> get corrupted files from any files transfer request.
> 
> -- David
> 
> >
> >
> > >
> > > -- David
> > >
> > > >
> > > > Andrew
> > > >
> >