Hello Alan,

The __ax88179_read_cmd() and __ax88179_write_cmd() routines are 
asynchronous with respect to ax88179_disconnect(), right?  Or at least, 
they are if they run as a result of the user closing the network 
interface.  Otherwise there wouldn't be any memory ordering issues.

Yes, I think so, they could be asynchronous regarding ax88179_disconnect.

But the memory barriers you added are not the proper solution.  What you 
need here is _synchronization_, not _ordering_.  As it is, the memory 
barriers you have added don't do anything; they shouldn't be in the 
patch.

Ok, thank you for the helpful clarification, let me check it better,
I understood it in a wrong way.

If you would like a more in-depth explanation, let me know.

Thank you for your help, I will try first, I really appreciate this.

Best regards
José Ignacio