On 6/3/25 08:03, lihuisong (C) wrote:

quoted

Once the memcpy_toio completes, the driver will not look at the 
packet again.  if the Kernel did change it at this point, it would 
not affect the flow.  The send of the packet is checked vi rc 
returned from send_data, and it tags the packet as dropped. Is this 
not sufficient?

Yes, it is not enough.
Once send_data() return success, platform can receive an interrupt，but 
the processing of the platform has not ended.
This processing includes handling data and then triggering an 
interrupt to notify OS. 

This comment caused me to rethink how I was using the PCC mailbox API.  
I realized that it was not actually enforcing the PCC protocol, which 
you identified.  It lead me to rewrite a postion of the PCC Mailbox API, 
and that is in my next patch series.  I would appreciate it if you would 
take a look. I think it addresses this concern, but it might not be 
completely transparent to a reviewer.   I would greatly appreciate if 
you were to look at it and confirm it fixes the issue, or, if I have 
missed something, let me know what you see.