9 messages in com.xensource.lists.xen-devel[Xen-devel] Re: [PATCH resend] allow ...
FromSent OnAttachments
Muli Ben-Yehuda26 Sep 2006 08:26 
Keir Fraser26 Sep 2006 08:50 
Muli Ben-Yehuda26 Sep 2006 09:08 
Keir Fraser26 Sep 2006 10:43 
Muli Ben-Yehuda26 Sep 2006 11:49 
Keir Fraser26 Sep 2006 12:46 
Keir Fraser27 Sep 2006 02:17 
Muli Ben-Yehuda27 Sep 2006 14:13 
Keir Fraser28 Sep 2006 01:42 
Subject:[Xen-devel] Re: [PATCH resend] allow connecting to xenconsole from remote hosts
From:Keir Fraser (Keir@cl.cam.ac.uk)
Date:09/27/2006 02:17:19 AM
List:com.xensource.lists.xen-devel

On 26/9/06 16:26, "Muli Ben-Yehuda" <mu@il.ibm.com> wrote:

The attached patch makes xenconsole send and receive console messages over a remote connection, instead of via stdin/stdout - if given the --remote switch. It is useful for proxy'ing real console message or other protocol messages (such as gdb) between dom0 and a remote host. We're currently using it for proxying gdb between gdbstub in a partition that talks gdb over the console page to a remote host running gdb.

Reconsidering this patch, I think it's okay to put socket connection capability into the client. *But*: 1. It can't go in until development reopens for 3.0.4. 2. I think the buffering code is not quite there yet. What you want is to make all fd's non-blocking, have small (e.g., 1kB) buffers, and then only add an fd to the read-set if its buffer is not full, and only add an fd to the write-set if its buffer is not empty. The buffers can stay as ring buffers, so a bunch of that code probably stays. 3. The networking options are confusing. What are the 'remote' and 'gateway' options intended for?

I think that the networking and buffering-fixes parts of the patch should be split. We can put in the buffering/select() changes first.

-- Keir