RE: [PATCH net-next v3 06/14] net: ethernet: oa_tc6: Support for hardware timestamp

[Selvamani Rajagopal]

> 
> Hi Rajagopal,
> 
> I have used your oa_tc6 framework to implement PTP support
> for the Microchip LAN8650. While working though this, I found
> two issues with the framework and a latent issue related to RX FCS.
> 
> 1. TTSCA capture base 0x1010 wrong. Should be 0x0010 Proven to work for LAN8650


Thanks for testing the PTP support and catching this. I am both intrigued and embarrassed. Intrigued because 
how/why it works in our setup with this address. I need to discuss with our hardware designer.

Irrespective, you are right and this definition is not following the specs. Will fix this in next, v4 submission


> 2. FTSE without FTSS leads to 32-bit prepend vs 8-byte strip, causing all RX
> to be corrupted -- CONFIG0 must be 0xC0.

I intentionally left it for vendor side code to initialize post oa_tc6_init because I thought it may be vendor dependent,
whether a vendor supports 32-bit timestamp or 64-bit timestamp,  Though I say this, oa_tc6.c reads 64 bits of 
timestamp data without checking this bit, before calling skb_tstamp_tx on the skb. So code is written for 64-bits 
timestamp.

I am open to suggestions whether I should move FTSS to oa_tc6.c 


> 3. RX FCS not stripped properly, resulting in ptp4l "bad message"

Expected unless vendor side code has 
       ndev->hw_features |= NETIF_F_RXFCS;
As you could see, 
1) oa_tc6.c checks for this bit before stripping the FCS.
2) Our code (onsemi/s2500.c) has this set and 

Hope this is ok. If not, open to suggestions. 

>