Xen devel - Sep 2006 - Not waiting for devices in xend when resuming a domain

Hello!

I noticed that if a domain is resumed it is NOT being waited for the 
hotplug scripts to finish running (call to waitForDevices in xend) , like 
this is for example done when a domain is starting. If I put a lot of load 
on domain-0 by for example compiling the sources with multiple threads 
(make -j 16) and do a resume on a domain, this can lead to access to 
devices although they are not properly connected, yet. I think it''s
even
more critical to have immediate and guaranteed access to all devices when 
a domain is immediately available (resume) and if it wants to access a 
device instead of it the case when it might be busy spending some time 
booting.

-- Stefan

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 23/9/06 5:31 pm, "Stefan Berger" <stefanb@us.ibm.com> wrote:
> I noticed that if a domain is resumed it is NOT being waited for the
hotplug
> scripts to finish running (call to waitForDevices in xend) , like this is
for
> example done when a domain is starting. If I put a lot of load on domain-0
by
> for example compiling the sources with multiple threads (make -j 16) and do
a
> resume on a domain, this can lead to access to devices although they are
not
> properly connected, yet. I think it''s even more critical to have
immediate and
> guaranteed access to all devices when a domain is immediately available
> (resume) and if it wants to access a device instead of it the case when it
> might be busy spending some time booting.
They at least need to be set up enough that the restored domain can start
connecting to its new backends. So the backend info directory needs to be
set up by the time the domain resumes execution.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Keir Fraser <Keir.Fraser@cl.cam.ac.uk> wrote on 09/23/2006 12:31:37 PM:
> 
> 
> 
> On 23/9/06 5:31 pm, "Stefan Berger" <stefanb@us.ibm.com>
wrote:
> I noticed that if a domain is resumed it is NOT being waited for the
> hotplug scripts to finish running (call to waitForDevices in xend) ,
> like this is for example done when a domain is starting. If I put a 
> lot of load on domain-0 by for example compiling the sources with 
> multiple threads (make -j 16) and do a resume on a domain, this can 
> lead to access to devices although they are not properly connected, 
> yet. I think it''s even more critical to have immediate and 
> guaranteed access to all devices when a domain is immediately 
> available (resume) and if it wants to access a device instead of it 
> the case when it might be busy spending some time booting. 
> 
> They at least need to be set up enough that the restored domain can 
> start connecting to its new backends. So the backend info directory 
> needs to be set up by the time the domain resumes execution.
> 
This one-line patch here syncs with the state of the hotplug scrtips 
before resuming the domain. It''s done as late as possible so in many
cases
no waiting for the scripts is actually necessary.

diff -r c4f3f719d997 tools/python/xen/xend/XendCheckpoint.py
--- a/tools/python/xen/xend/XendCheckpoint.py            Sat Sep 23 
13:54:58 2006
+++ b/tools/python/xen/xend/XendCheckpoint.py            Sat Sep 23 
15:10:17 2006
@@ -163,6 +163,7 @@
 
         #Block until src closes connection
         os.read(fd, 1)
+        dominfo.waitForDevices()
         dominfo.unpause()
 
         dominfo.completeRestore(handler.store_mfn, handler.console_mfn)

>  -- Keir
  Stefan


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel