On Thu, Apr 8, 2010 at 7:09 PM, Ben Pfaff <span dir="ltr"><<a href="mailto:blp@nicira.com">blp@nicira.com</a>></span> wrote:<br><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
I pushed a small change to the netdev library to the "wdp" branch this<br>
morning. Since this is not a "core" branch (yet) and already has a<br>
bunch of badly broken stuff in it, I did not see a need to have it<br>
reviewed in advance, but it seems like a good idea to get it reviewed<br>
now. Here's the commit.<br></blockquote><div><br></div><div>The code looks correct to me, though what do you think about calling it something like netdev_clone() instead? netdev_reopen() isn't the most descriptive name to me.</div>
<div><br></div><div>I'm assuming that these netdevs are just used to get properties of devices and won't be used to send and receive? Obviously since they are are just refcounted they share file descriptors.</div>
<div><br></div><div>Finally, I haven't really looked at the wdp branch so I don't know the context but do you think the performance benefits are worth it? It took me a while to understand all the levels to which struct sk_buff in Linux can be shared or independent. This isn't nearly as bad but this is now the second refcount that we have on netdevs. How does this compare to struct wdp_port that contains the netdev?</div>
</div>