Nithin, in a NEW NL command, when the WFP filter needs to be created for a new 
tunnel port, the code runs at dispatch IRQL. At dispatch IRQL any WFP call 
fails. That is why I have created a thread for postponing WFP-based work when 
the IRQL is lowered at passive level and the WFP calls can be successfully 
executed.

I am not in favor of temporarily lowering the IRQL, because this will break the 
purpose of mutual exclusion that is used for the switch ports.

-----Original Message-----
From: Nithin Raju [mailto:[email protected]] 
Sent: Tuesday, 27 January, 2015 20:01
To: Eitan Eliahu
Cc: Sorin Vinturis; [email protected]
Subject: Re: [ovs-dev] [PATCH v2] datapath-windows: Support for custom VXLAN 
tunnel port

Eitan,
Like we discussed offline, I am not comfortable with a netlink call waiting on 
another thread in a synchronous context. For what it is worth, all the calls 
from userspace to kernel are non-blocking. It would be great if we can not have 
a separate thread, and instead use some primitives such as atomic variables to 
synchronize between the WFP calls. We can talk more during the IRC meeting.

thanks,
-- Nithin

> On Jan 26, 2015, at 1:34 PM, Eitan Eliahu <[email protected]> wrote:
> 
> 
> Hi Sorin,
> Thank you for addressing these issues. Please find some minor comments below.
> The issue I see now is that the IOCTL completion is not synchronized with the 
> engine execution and if the transaction fails it is not propagated to user 
> mode . I realize that the thread exists when the transaction fails but we 
> might just fail the transaction.
> I think once we have the IOCTL synced we can go ahead with this challenging 
> change.
> Eitan

_______________________________________________
dev mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/dev

Reply via email to