On Thu, Aug 25, 2011 at 02:27:56AM +0000, KY Srinivasan wrote:

Since I don't have any (current) use for the driver_data pointer, I have gone ahead
and cleaned up the first 74 patches without adding the driver_data. 
With the mushing of the patches   you had proposed this is about
a 60 patch series and addresses all the other comments you had in the first 74 patches.
I hope I have gotten the "right" granularity now.   If it is ok with you, I could send these 
out for your consideration.

Please do.

But if you do, do you mind if I add the driver_data pointer, so you can
blame me later if no one uses it?  :)

The only unresolved issue in the remaining patches (75 - 117) is the reference counting
issue we have been debating. As I noted in my earlier emails on this topic, the reference
counting has been there for a long time and I am reluctant get rid of that code without 
additional testing/analysis. So I want to propose the following options:

1) Keep the existing code and I will skip the patches that cleaned up the reference counting

2) Take the cleanup that I have implemented

In either case, I would further test and analyze this code to see if (a) the race condition that is being
addressed is valid and (b) if there is a different mechanism that could be used to deal with it. Given
the gaping holes in the current implementation, my personal preference would be to go with the 
second option. Let me know what you want me to do here.  

Ok, that sounds acceptable, but don't add the lock to the hv_driver, or
is that needed right now?

thanks,

greg k-h